development and implementation of slam algorithms
play

Development and Implementation of SLAM Algorithms Kasra Khosoussi - PowerPoint PPT Presentation

Jan 26, 2024 •571 likes •2.05k views

Development and Implementation of SLAM Algorithms Kasra Khosoussi Supervised by: Dr. Hamid D. Taghirad Advanced Robotics and Automated Systems (ARAS) Industrial Control Laboratory K.N. Toosi University of Technology July 13, 2011 K. Khosoussi

  1. Introduction Bayesian Filtering Probabilistic Methods Probabilistic methods outperform deterministic algorithms Describe uncertainty in data, models Filtering distribution: and estimates p ( s t , θ | z 1: t , u 1: t ) Bayesian estimation Smoothing distribution: p ( s 0 : t , θ | z 1: t , u 1: t ) - Robot pose s t MMSE estimate: - Robot pose is assumed to be a Markov x t = E [ x t | z 1: t , u 1: t ] ˆ process with initial distribution p ( s 0 ) x 0: t = E [ x 0: t | z 1: t , u 1: t ] ˆ - Feature’s location θ i - Map θ = { θ 1 , . . . , θ N } - Observation z t , and control input u t θ ] T - x t = [ s t - x 1: t � { x 1 , . . . , x t } K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 7 / 43

  2. Introduction Bayesian Filtering Probabilistic Methods Probabilistic methods outperform deterministic algorithms Describe uncertainty in data, models Filtering distribution: and estimates p ( s t , θ | z 1: t , u 1: t ) Bayesian estimation Smoothing distribution: p ( s 0 : t , θ | z 1: t , u 1: t ) - Robot pose s t MMSE estimate: - Robot pose is assumed to be a Markov x t = E [ x t | z 1: t , u 1: t ] ˆ process with initial distribution p ( s 0 ) x 0: t = E [ x 0: t | z 1: t , u 1: t ] ˆ - Feature’s location θ i - Map θ = { θ 1 , . . . , θ N } - Observation z t , and control input u t θ ] T - x t = [ s t - x 1: t � { x 1 , . . . , x t } K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 7 / 43

  3. Introduction Bayesian Filtering State-Space Equations Robot motion equation: s t = f ( s t − 1 , u t , v t ) Observation equation: z t = g ( s t , θ n t , w t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 8 / 43

  4. Introduction Bayesian Filtering State-Space Equations Robot motion equation: s t = f ( s t − 1 , u t , v t ) Observation equation: z t = g ( s t , θ n t , w t ) f ( · , · , · ) and g ( · , · , · ) are non-linear functions K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 8 / 43

  5. Introduction Bayesian Filtering State-Space Equations Robot motion equation: s t = f ( s t − 1 , u t , v t ) Observation equation: z t = g ( s t , θ n t , w t ) f ( · , · , · ) and g ( · , · , · ) are non-linear functions v t and w t are zero-mean white Gaussian noises with covariances matrices Q t and R t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 8 / 43

  6. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  7. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  8. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  9. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  10. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  11. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  12. Introduction Bayesian Filtering Bayes Filter How to estimate the posterior distribution recursively in time? Bayes filter! 1 Prediction 2 Update � p ( x t | z 1: t − 1 , u 1: t ) = p ( x t | x t − 1 , u t ) p ( x t − 1 | z 1: t − 1 , u 1: t − 1 ) d x t − 1 (1) p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) p ( x t | z 1: t , u 1: t ) = (2) � p ( z t | x t ) p ( x t | z 1: t − 1 , u 1: t ) d x t Motion model Observation model There is a similar recursive formula for the smoothing density K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 9 / 43

  13. Introduction Bayesian Filtering Bayes Filter Cont’d. Example For the case of linear-Gaussian models, Bayes filter equations would be simplified into the Kalman filter equations. K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 10 / 43

  14. Introduction Bayesian Filtering Bayes Filter Cont’d. Example For the case of linear-Gaussian models, Bayes filter equations would be simplified into the Kalman filter equations. But . . . In general, it is impossible to implement the exact Bayes filter because it requires the ability to evaluate complex high-dimensional integrals. K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 10 / 43

  15. Introduction Bayesian Filtering Bayes Filter Cont’d. Example For the case of linear-Gaussian models, Bayes filter equations would be simplified into the Kalman filter equations. But . . . In general, it is impossible to implement the exact Bayes filter because it requires the ability to evaluate complex high-dimensional integrals. So we have to use approximation . . . Extended Kalman Filter (EKF) Unscented Kalman Filter (UKF) Gaussian-Sum Filter Extended Information Filter (EIF) Particle Filter (A.K.A. Sequential Monte Carlo Methods) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 10 / 43

  16. Introduction Bayesian Filtering Bayes Filter Cont’d. Example For the case of linear-Gaussian models, Bayes filter equations would be simplified into the Kalman filter equations. But . . . In general, it is impossible to implement the exact Bayes filter because it requires the ability to evaluate complex high-dimensional integrals. So we have to use approximation . . . Extended Kalman Filter (EKF) Unscented Kalman Filter (UKF) Gaussian-Sum Filter Extended Information Filter (EIF) Particle Filter (A.K.A. Sequential Monte Carlo Methods) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 10 / 43

  17. Introduction Particle Filter Perfect Monte Carlo Sampling � Q. How to compute expected values such as E p ( x ) [ h ( x )] = h ( x ) p ( x ) d x for any integrable function h ( · ) ? K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 11 / 43

  18. Introduction Particle Filter Perfect Monte Carlo Sampling � Q. How to compute expected values such as E p ( x ) [ h ( x )] = h ( x ) p ( x ) d x for any integrable function h ( · ) ? Perfect Monte Carlo (A.K.A. Monte Carlo Integration) 1 Generate N i.i.d. samples { x [ i ] } N i =1 according to p ( x ) 2 Estimate the PDF as P N ( x ) � 1 � N i =1 δ ( x − x [ i ] ) N h ( x ) P N ( x ) d x = 1 � N 3 Estimate E p ( x ) [ h ( x )] ≈ i =1 h ( x [ i ] ) � N K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 11 / 43

  19. Introduction Particle Filter Perfect Monte Carlo Sampling � Q. How to compute expected values such as E p ( x ) [ h ( x )] = h ( x ) p ( x ) d x for any integrable function h ( · ) ? Perfect Monte Carlo (A.K.A. Monte Carlo Integration) 1 Generate N i.i.d. samples { x [ i ] } N i =1 according to p ( x ) 2 Estimate the PDF as P N ( x ) � 1 � N i =1 δ ( x − x [ i ] ) N h ( x ) P N ( x ) d x = 1 � N 3 Estimate E p ( x ) [ h ( x )] ≈ i =1 h ( x [ i ] ) � N Convergence theorems for N → ∞ using central limit theorem and strong law of large numbers Error decreases with O ( N − 1 / 2 ) regardless of the dimension of x K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 11 / 43

  20. Introduction Particle Filter Perfect Monte Carlo Sampling � Q. How to compute expected values such as E p ( x ) [ h ( x )] = h ( x ) p ( x ) d x for any integrable function h ( · ) ? Perfect Monte Carlo (A.K.A. Monte Carlo Integration) 1 Generate N i.i.d. samples { x [ i ] } N i =1 according to p ( x ) 2 Estimate the PDF as P N ( x ) � 1 � N i =1 δ ( x − x [ i ] ) N h ( x ) P N ( x ) d x = 1 � N 3 Estimate E p ( x ) [ h ( x )] ≈ i =1 h ( x [ i ] ) � N Convergence theorems for N → ∞ using central limit theorem and strong law of large numbers Error decreases with O ( N − 1 / 2 ) regardless of the dimension of x But . . . It is usually impossible to sample directly from the filtering or smoothing distribution (high-dimensional, non-standard, only known up to a constant) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 11 / 43

  21. Introduction Particle Filter Importance Sampling (IS) Idea Generate samples from another distribution called the importance function like π ( x ) , and weight these samples according to w ∗ ( x [ i ] ) = p ( x [ i ] ) π ( x [ i ] ) : K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 12 / 43

  22. Introduction Particle Filter Importance Sampling (IS) Idea Generate samples from another distribution called the importance function like π ( x ) , and weight these samples according to w ∗ ( x [ i ] ) = p ( x [ i ] ) π ( x [ i ] ) : N π ( x ) π ( x ) d x = 1 � h ( x ) p ( x ) � w ∗ ( x [ i ] ) h ( x [ i ] ) E p ( x ) [ h ( x )] = N i =1 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 12 / 43

  23. Introduction Particle Filter Importance Sampling (IS) Idea Generate samples from another distribution called the importance function like π ( x ) , and weight these samples according to w ∗ ( x [ i ] ) = p ( x [ i ] ) π ( x [ i ] ) : N π ( x ) π ( x ) d x = 1 � h ( x ) p ( x ) � w ∗ ( x [ i ] ) h ( x [ i ] ) E p ( x ) [ h ( x )] = N i =1 But . . . How to do this recursively in time? K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 12 / 43

  24. Introduction Particle Filter Sequential Importance Sampling (SIS) Sampling from scratch from the importance function π ( x 0: t | z 1: t , u 1: t ) implies growing computational complexity for each step over time K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 13 / 43

  25. Introduction Particle Filter Sequential Importance Sampling (SIS) Sampling from scratch from the importance function π ( x 0: t | z 1: t , u 1: t ) implies growing computational complexity for each step over time Q. How to estimate p ( x 0: t | z 1: t , u 1: t ) using importance sampling recursively? K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 13 / 43

  26. Introduction Particle Filter Sequential Importance Sampling (SIS) Sampling from scratch from the importance function π ( x 0: t | z 1: t , u 1: t ) implies growing computational complexity for each step over time Q. How to estimate p ( x 0: t | z 1: t , u 1: t ) using importance sampling recursively? Sequential Importance Sampling At time t , generate x [ i ] t according to π ( x t | x [ i ] 0: t − 1 , z 1: t , u 1: t ) (proposal distribution), and merge it with the previous samples x [ i ] 0: t − 1 drawn from π ( x 0: t − 1 | z 1: t − 1 , u 1: t − 1 ) : x [ i ] 0: t = { x [ i ] 0: t − 1 , x [ i ] t } ∼ π ( x 0: t | z 1: t , u 1: t ) 0: t − 1 ) p ( z t | x [ i ] t ) p ( x [ i ] t | x [ i ] t − 1 , u t ) w ( x [ i ] 0: t ) = w ( x [ i ] π ( x [ i ] t | x [ i ] 0: t − 1 , z 1: t , u 1: t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 13 / 43

  27. Introduction Particle Filter Degeneracy and Resampling Degeneracy Problem After a few steps, all but one of the particles (samples) would have very insignificant normalized weight K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 14 / 43

  28. Introduction Particle Filter Degeneracy and Resampling Degeneracy Problem After a few steps, all but one of the particles (samples) would have very insignificant normalized weight Resampling Eliminate particles with low normalized weights and multiply those with high normalized weights in a probabilistic manner K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 14 / 43

  29. Introduction Particle Filter Degeneracy and Resampling Degeneracy Problem After a few steps, all but one of the particles (samples) would have very insignificant normalized weight Resampling Eliminate particles with low normalized weights and multiply those with high normalized weights in a probabilistic manner Resampling will cause sample impoverishment K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 14 / 43

  30. Introduction Particle Filter Degeneracy and Resampling Degeneracy Problem After a few steps, all but one of the particles (samples) would have very insignificant normalized weight Resampling Eliminate particles with low normalized weights and multiply those with high normalized weights in a probabilistic manner Resampling will cause sample impoverishment Effective sample size (ESS) is a measure of the degeneracy of SIS that can be used in order to avoid unnecessary resampling steps 1 ˆ N eff = w ( x [ i ] � N 0: t ) 2 i =1 ˜ Perform resampling only if N eff is lower than a fixed threshold N T K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 14 / 43

  31. Introduction Particle Filter Proposal Distribution Selecting an appropriate proposal distribution π ( x t | x 0: t − 1 , z 1: t , u 1: t ) plays an important role in the success of particle filter The simplest and most common choice is the motion model (transition density) p ( x t | x t − 1 , u t ) p ( x t | x t − 1 , z t , u t ) is known as the optimal proposal distribution and limits the degeneracy of the particle filter by minimizing the conditional variance of unnormalized weights Importance weights for the optimal proposal distribution can be obtained as: w ( x [ i ] 0: t ) = w ( x [ i ] 0: t − 1 ) p ( z t | x [ i ] t − 1 ) But . . . In SLAM, neither p ( x t | x t − 1 , z t , u t ) nor p ( z t | x [ i ] t − 1 ) can be computed in closed form and we have to use approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 15 / 43

  32. Introduction Particle Filter Proposal Distribution Selecting an appropriate proposal distribution π ( x t | x 0: t − 1 , z 1: t , u 1: t ) plays an important role in the success of particle filter The simplest and most common choice is the motion model (transition density) p ( x t | x t − 1 , u t ) p ( x t | x t − 1 , z t , u t ) is known as the optimal proposal distribution and limits the degeneracy of the particle filter by minimizing the conditional variance of unnormalized weights Importance weights for the optimal proposal distribution can be obtained as: w ( x [ i ] 0: t ) = w ( x [ i ] 0: t − 1 ) p ( z t | x [ i ] t − 1 ) But . . . In SLAM, neither p ( x t | x t − 1 , z t , u t ) nor p ( z t | x [ i ] t − 1 ) can be computed in closed form and we have to use approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 15 / 43

  33. Introduction Particle Filter Proposal Distribution Selecting an appropriate proposal distribution π ( x t | x 0: t − 1 , z 1: t , u 1: t ) plays an important role in the success of particle filter The simplest and most common choice is the motion model (transition density) p ( x t | x t − 1 , u t ) p ( x t | x t − 1 , z t , u t ) is known as the optimal proposal distribution and limits the degeneracy of the particle filter by minimizing the conditional variance of unnormalized weights Importance weights for the optimal proposal distribution can be obtained as: w ( x [ i ] 0: t ) = w ( x [ i ] 0: t − 1 ) p ( z t | x [ i ] t − 1 ) But . . . In SLAM, neither p ( x t | x t − 1 , z t , u t ) nor p ( z t | x [ i ] t − 1 ) can be computed in closed form and we have to use approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 15 / 43

  34. Introduction Particle Filter Proposal Distribution Selecting an appropriate proposal distribution π ( x t | x 0: t − 1 , z 1: t , u 1: t ) plays an important role in the success of particle filter The simplest and most common choice is the motion model (transition density) p ( x t | x t − 1 , u t ) p ( x t | x t − 1 , z t , u t ) is known as the optimal proposal distribution and limits the degeneracy of the particle filter by minimizing the conditional variance of unnormalized weights Importance weights for the optimal proposal distribution can be obtained as: w ( x [ i ] 0: t ) = w ( x [ i ] 0: t − 1 ) p ( z t | x [ i ] t − 1 ) But . . . In SLAM, neither p ( x t | x t − 1 , z t , u t ) nor p ( z t | x [ i ] t − 1 ) can be computed in closed form and we have to use approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 15 / 43

  35. Introduction Particle Filter Proposal Distribution Selecting an appropriate proposal distribution π ( x t | x 0: t − 1 , z 1: t , u 1: t ) plays an important role in the success of particle filter The simplest and most common choice is the motion model (transition density) p ( x t | x t − 1 , u t ) p ( x t | x t − 1 , z t , u t ) is known as the optimal proposal distribution and limits the degeneracy of the particle filter by minimizing the conditional variance of unnormalized weights Importance weights for the optimal proposal distribution can be obtained as: w ( x [ i ] 0: t ) = w ( x [ i ] 0: t − 1 ) p ( z t | x [ i ] t − 1 ) But . . . In SLAM, neither p ( x t | x t − 1 , z t , u t ) nor p ( z t | x [ i ] t − 1 ) can be computed in closed form and we have to use approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 15 / 43

  36. RBPF-SLAM Table of Contents Introduction 1 SLAM Problem Bayesian Filtering Particle Filter RBPF-SLAM 2 Monte Carlo Approximation of the Optimal Proposal Distribution 3 Introduction LRS LIS-1 LIS-2 Results 4 Simulation Results Experiments On Real Data Conclusion 5 Future Work 6 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 16 / 43

  37. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  38. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  39. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  40. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  41. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  42. RBPF-SLAM Rao-Blackwellized Particle Filter in SLAM SLAM is a very high-dimensional problem Estimating the p ( s 0: t , θ | z 1: t , u 1: t ) using a particle filter can be very inefficient We can factor the smoothing distribution into two parts as M � p ( s 0: t , θ | z 1: t , u 1: t ) = p ( s 0: t | z 1: t , u 1: t ) p ( θ k | s 0: t , z 1: t , u 1: t ) k =1 Motion model is used as the proposal distribution in FastSLAM 1.0 FastSLAM 2.0 linearizes the observation equation and approximates the optimal proposal distribution with a Gaussian distribution FastSLAM 2.0 outperforms FastSLAM 1.0 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 17 / 43

  43. RBPF-SLAM FastSLAM 2.0 Linearization error K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 18 / 43

  44. RBPF-SLAM FastSLAM 2.0 Linearization error Gaussian approximation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 18 / 43

  45. RBPF-SLAM FastSLAM 2.0 Linearization error Gaussian approximation Linear motion models with respect to the noise variable v t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 18 / 43

  46. Monte Carlo Approximation of the Optimal Proposal Distribution Table of Contents Introduction 1 SLAM Problem Bayesian Filtering Particle Filter RBPF-SLAM 2 Monte Carlo Approximation of the Optimal Proposal Distribution 3 Introduction LRS LIS-1 LIS-2 Results 4 Simulation Results Experiments On Real Data Conclusion 5 Future Work 6 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 19 / 43

  47. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  48. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  49. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  50. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation 4 Perform resampling K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  51. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation 4 Perform resampling Proposed Algorithms 1 Sample from the optimal proposal distribution using Monte Carlo sampling methods K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  52. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation 4 Perform resampling Proposed Algorithms 1 Sample from the optimal proposal distribution using Monte Carlo sampling methods 2 Update the landmarks EKFs for the observed features K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  53. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation 4 Perform resampling Proposed Algorithms 1 Sample from the optimal proposal distribution using Monte Carlo sampling methods 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using Monte Carlo integration K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  54. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction FastSLAM 2.0 v.s. The Proposed Algorithms FastSLAM 2.0 1 Approximate the optimal proposal distribution with a Gaussian using the linearized observation equation. Sample from this Gaussian 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using the linearized observation equation 4 Perform resampling Proposed Algorithms 1 Sample from the optimal proposal distribution using Monte Carlo sampling methods 2 Update the landmarks EKFs for the observed features 3 Compute the importance weights using Monte Carlo integration 4 Perform resampling only if it is necessary according to ESS K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 20 / 43

  55. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  56. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) Instead of sampling directly from the optimal proposal distribution . . . We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and then weight those samples p ( s t | s [ i ] t − 1 , z t , u t ) proportional to q ( s t | s [ i ] t − 1 , z t , u t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  57. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) Instead of sampling directly from the optimal proposal distribution . . . We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and then weight those samples p ( s t | s [ i ] t − 1 , z t , u t ) proportional to t − 1 , z t , u t ) Local Importance Sampling (LIS) q ( s t | s [ i ] K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  58. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) Instead of sampling directly from the optimal proposal distribution . . . We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and then weight those samples p ( s t | s [ i ] t − 1 , z t , u t ) proportional to t − 1 , z t , u t ) Local Importance Sampling (LIS) q ( s t | s [ i ] We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and accept some of them as samples of the optimal proposal distribution according to rejection sampling criteria K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  59. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) Instead of sampling directly from the optimal proposal distribution . . . We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and then weight those samples p ( s t | s [ i ] t − 1 , z t , u t ) proportional to t − 1 , z t , u t ) Local Importance Sampling (LIS) q ( s t | s [ i ] We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and accept some of them as samples of the optimal proposal distribution according to rejection sampling criteria Local Rejection Sampling (LRS) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  60. Monte Carlo Approximation of the Optimal Proposal Distribution Introduction MC Approximation of The Optimal Proposal Dist. MC sampling methods such as importance sampling (IS) and rejection sampling (RS) can be used in order to sample from the optimal proposal distribution p ( s t | s [ i ] t − 1 , z t , u t ) Instead of sampling directly from the optimal proposal distribution . . . We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and then weight those samples p ( s t | s [ i ] t − 1 , z t , u t ) proportional to t − 1 , z t , u t ) Local Importance Sampling (LIS) q ( s t | s [ i ] We can generate M samples { s [ i,j ] } M j =1 according to another t distribution like q ( s t | s [ i ] t − 1 , z t , u t ) and accept some of them as samples of the optimal proposal distribution according to rejection sampling criteria Local Rejection Sampling (LRS) q ( s t | s [ i ] t − 1 , u t , z t ) = p ( s t | s [ i ] t − 1 , u t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 21 / 43

  61. Monte Carlo Approximation of the Optimal Proposal Distribution LRS Local Rejection Sampling (LRS) graphics by M. Jordan K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 22 / 43

  62. Monte Carlo Approximation of the Optimal Proposal Distribution LRS Local Rejection Sampling (LRS) graphics by M. Jordan Rejection Sampling 1 Generate u [ i ] ∼ U [0 , 1] and s [ i,j ] ∼ p ( s t | s [ i ] t − 1 , u t ) t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 22 / 43

  63. Monte Carlo Approximation of the Optimal Proposal Distribution LRS Local Rejection Sampling (LRS) graphics by M. Jordan Rejection Sampling 1 Generate u [ i ] ∼ U [0 , 1] and s [ i,j ] ∼ p ( s t | s [ i ] t − 1 , u t ) t p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Accept s [ i,j ] if u [ i ] ≤ t t C · p ( s [ i,j ] | s [ i ] t − 1 , u t ) t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 22 / 43

  64. Monte Carlo Approximation of the Optimal Proposal Distribution LRS Local Rejection Sampling (LRS) graphics by M. Jordan Rejection Sampling 1 Generate u [ i ] ∼ U [0 , 1] and s [ i,j ] ∼ p ( s t | s [ i ] t − 1 , u t ) t p ( s [ i,j ] | s [ i ] p ( z t | s [ i,j ] t − 1 , z t , u t ) 2 Accept s [ i,j ] if u [ i ] ≤ ) t t − 1 , u t ) = t t C · p ( s [ i,j ] | s [ i ] p ( z t | s [ i,j ] max ) t t j K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 22 / 43

  65. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  66. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples We should compute p ( z t | s [ i ] t − 1 ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  67. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples We should compute p ( z t | s [ i ] t − 1 ) M 1 � p ( z t | s [ i ] p ( z t | s t ) p ( s t | s [ i ] p ( z t | s [ i,j ] � t − 1 ) = t − 1 , u t ) d s t ≈ ) t M MC Intergration j =1 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  68. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples We should compute p ( z t | s [ i ] t − 1 ) M 1 � p ( z t | s [ i ] p ( z t | s t ) p ( s t | s [ i ] p ( z t | s [ i,j ] � t − 1 ) = t − 1 , u t ) d s t ≈ ) t M MC Intergration j =1 p ( z t | s [ i,j ] ) can be approximated by a Gaussian t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  69. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples We should compute p ( z t | s [ i ] t − 1 ) M 1 � p ( z t | s [ i ] p ( z t | s t ) p ( s t | s [ i ] p ( z t | s [ i,j ] � t − 1 ) = t − 1 , u t ) d s t ≈ ) t M MC Intergration j =1 p ( z t | s [ i,j ] ) can be approximated by a Gaussian t MC Integration K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  70. Monte Carlo Approximation of the Optimal Proposal Distribution LRS LRS Cont’d. Now we have to compute the importance weights for the set of accepted samples We should compute p ( z t | s [ i ] t − 1 ) M 1 � p ( z t | s [ i ] p ( z t | s t ) p ( s t | s [ i ] p ( z t | s [ i,j ] � t − 1 ) = t − 1 , u t ) d s t ≈ ) t M MC Intergration j =1 p ( z t | s [ i,j ] ) can be approximated by a Gaussian t MC Integration ⇒ Large number of local particles M K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 23 / 43

  71. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  72. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Local IS weights: w ( LIS ) ( s [ i,j ] ) = p ( z t | s [ i,j ] ) ∝ t t t p ( s [ i,j ] | s [ i ] t − 1 , u t ) t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  73. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Local IS weights: w ( LIS ) ( s [ i,j ] ) = p ( z t | s [ i,j ] ) ∝ t t t p ( s [ i,j ] | s [ i ] t − 1 , u t ) t 3 Local resampling among { s [ i,j ] j =1 using w ( LIS ) ( s [ i,j ] } M ) t t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  74. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Local IS weights: w ( LIS ) ( s [ i,j ] ) = p ( z t | s [ i,j ] ) ∝ t t t p ( s [ i,j ] | s [ i ] t − 1 , u t ) t 3 Local resampling among { s [ i,j ] j =1 using w ( LIS ) ( s [ i,j ] } M ) t t 4 Main weights: w ( s ∗ [ i,j ] ) = w ( s [ i ] t − 1 ) p ( z t | s [ i ] t − 1 ) t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  75. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Local IS weights: w ( LIS ) ( s [ i,j ] ) = p ( z t | s [ i,j ] ) ∝ t t t p ( s [ i,j ] | s [ i ] t − 1 , u t ) t 3 Local resampling among { s [ i,j ] j =1 using w ( LIS ) ( s [ i,j ] } M ) t t 4 Main weights: w ( s ∗ [ i,j ] ) = w ( s [ i ] t − 1 ) p ( z t | s [ i ] t − 1 ) t 5 p ( z t | s [ i ] � M j =1 p ( z t | s [ i,j ] 1 t − 1 ) ≈ ) t M MC Integration K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  76. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-1 Local Importance Sampling (LIS) LIS-1 LIS-1 1 Similar to LRS, we have to generate M random samples { s [ i,j ] } M t j =1 according to p ( s t | s [ i ] t − 1 , u t ) p ( s [ i,j ] | s [ i ] t − 1 , z t , u t ) 2 Local IS weights: w ( LIS ) ( s [ i,j ] ) = p ( z t | s [ i,j ] ) ∝ t t t p ( s [ i,j ] | s [ i ] t − 1 , u t ) t 3 Local resampling among { s [ i,j ] j =1 using w ( LIS ) ( s [ i,j ] } M ) t t 4 Main weights: w ( s ∗ [ i,j ] ) = w ( s [ i ] t − 1 ) p ( z t | s [ i ] t − 1 ) t 5 p ( z t | s [ i ] � M j =1 p ( z t | s [ i,j ] 1 t − 1 ) ≈ ) t M MC Integration K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 24 / 43

  77. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-2 Local Importance Sampling (LIS) LIS-2 LIS-2 Similar to LRS and LIS-1, we have to generate M random samples { s [ i,j ] j =1 according to p ( s t | s [ i ] } M t − 1 , u t ) t K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 25 / 43

  78. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-2 Local Importance Sampling (LIS) LIS-2 LIS-2 Similar to LRS and LIS-1, we have to generate M random samples { s [ i,j ] j =1 according to p ( s t | s [ i ] } M t − 1 , u t ) t Total weights of generated particles { s [ i,j ] } N,M ( i,j )=(1 , 1) = local weights t × SIS weights K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 25 / 43

  79. Monte Carlo Approximation of the Optimal Proposal Distribution LIS-2 Local Importance Sampling (LIS) LIS-2 LIS-2 Similar to LRS and LIS-1, we have to generate M random samples { s [ i,j ] j =1 according to p ( s t | s [ i ] } M t − 1 , u t ) t Total weights of generated particles { s [ i,j ] } N,M ( i,j )=(1 , 1) = local weights t × SIS weights Instead of eliminating local weights through local resamplings (LIS-1), total weights are computed in LIS-2 K. Khosoussi (ARAS) Development of SLAM Algorithms July 13, 2011 25 / 43

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

SLAM: COMPARATIVE APPROACH Khooshal Saurty 1 OUTLINE Introduction - What is SLAM? EKF SLAM

SLAM: COMPARATIVE APPROACH Khooshal Saurty 1 OUTLINE Introduction - What is SLAM? EKF SLAM

Intelligent Robotics Seminar - 31 October 2016 SLAM: COMPARATIVE APPROACH Khooshal Saurty 1 OUTLINE Introduction - What is SLAM? EKF SLAM FAST SLAM Comparison Cartographer Conclusion and References 2 INTRODUCTION - WHAT IS SLAM?

211 views • 20 slides
SLAM Tutorial Course 6.834 Cognitive Robotics SLAM for Dummies Authors: Morten Rufus Blas,

SLAM Tutorial Course 6.834 Cognitive Robotics SLAM for Dummies Authors: Morten Rufus Blas,

SLAM Tutorial Course 6.834 Cognitive Robotics SLAM for Dummies Authors: Morten Rufus Blas, Soren Riisgaard, May 2 0 0 4 Outline Project background Goal for the project Previous work Progress Implementation Screenshots

166 views • 5 slides
SLAM@NVIDIA Kari Pulli| Senior Director of Research Overview Keyframe-based SlAM 3D

SLAM@NVIDIA Kari Pulli| Senior Director of Research Overview Keyframe-based SlAM 3D

SLAM@NVIDIA Kari Pulli| Senior Director of Research Overview Keyframe-based SlAM 3D rendering for Augmented Reality Problems with traditional keyframe-based SLAM Solution: Deferred Triangulation SLAM KeyFrame-based SLAM 3D

724 views • 35 slides
I ntroduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard 1 The SLAM

I ntroduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard 1 The SLAM

I ntroduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard 1 The SLAM Problem SLAM stands for simultaneous localization and mapping The task of building a map while estimating the pose of the robot relative to

830 views • 43 slides
Visual SLAM for Mobile Instructor - Simon Lucey 16-423 - Designing Computer Vision Apps Example

Visual SLAM for Mobile Instructor - Simon Lucey 16-423 - Designing Computer Vision Apps Example

Visual SLAM for Mobile Instructor - Simon Lucey 16-423 - Designing Computer Vision Apps Example of SLAM for AR Taken from: H. Liu et al. Robust Keyframe-based Monocular SLAM for Augmented Reality, ISMAR 2016. Example of SLAM for AR

1.96k views • 55 slides
Visual SLAM for Mobile Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Example

Visual SLAM for Mobile Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Example

Visual SLAM for Mobile Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Example of SLAM for AR Taken from: H. Liu et al. Robust Keyframe-based Monocular SLAM for Augmented Reality, ISMAR 2016. Example of SLAM for AR

571 views • 55 slides
Direct Visual SLAM Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Reminder:

Direct Visual SLAM Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Reminder:

Direct Visual SLAM Instructor - Simon Lucey 16-623 - Designing Computer Vision Apps Reminder: SLAM S imultaneous L ocalization a nd M apping. On mobile interested primarily in Visual SLAM (VSLAM). Sometimes called Mono SLAM if there

617 views • 61 slides
Localisation, Mapping and the Simultaneous Localisation and Mapping (SLAM) Problem Hugh

Localisation, Mapping and the Simultaneous Localisation and Mapping (SLAM) Problem Hugh

Localisation, Mapping and the Simultaneous Localisation and Mapping (SLAM) Problem Hugh Durrant-Whyte Australian Centre for Field Robotics The University of Sydney hugh@acfr.usyd.edu.au SLAM Summer School 2002 Slide 1 Introduction SLAM

865 views • 49 slides
Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard, Michael Ruhnke, Bastian Steder What is SLAM? Estimate the pose of a robot and the map of the environment at the same time SLAM is hard,

743 views • 49 slides
I ntroduction to Mobile Robotics SLAM Landm ark-based FastSLAM Wolfram Burgard Partial

I ntroduction to Mobile Robotics SLAM Landm ark-based FastSLAM Wolfram Burgard Partial

I ntroduction to Mobile Robotics SLAM Landm ark-based FastSLAM Wolfram Burgard Partial slide courtesy of Mike Montemerlo 1 The SLAM Problem SLAM stands for simultaneous localization and mapping The task of building a map while

637 views • 36 slides
Shane Lawtey, SLAM Team Leader Events Professional Oweek Development Harmony Day

Shane Lawtey, SLAM Team Leader Events Professional Oweek Development Harmony Day

Presenter: Shane Lawtey, SLAM Team Leader Events Professional Oweek Development Harmony Day Career Ready Program Program Move in May Good Health Cover letter and resume Month Great Idea! writing workshop Spring

580 views • 28 slides
SLAM Landmark-based FastSLAM Wolfram Burgard, Maren Bennewitz, Diego Tipaldi, Luciano

SLAM Landmark-based FastSLAM Wolfram Burgard, Maren Bennewitz, Diego Tipaldi, Luciano

Introduction to Mobile Robotics SLAM Landmark-based FastSLAM Wolfram Burgard, Maren Bennewitz, Diego Tipaldi, Luciano Spinello Partial slide courtesy of Mike Montemerlo 1 The SLAM Problem SLAM stands for simultaneous localization and

361 views • 32 slides
Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Cyrill Stachniss,

Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Cyrill Stachniss,

Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Cyrill Stachniss, Maren Bennewitz, Diego Tipaldi, Luciano Spinello 1 The SLAM Problem SLAM stands for simultaneous localization and mapping The task of

1.14k views • 42 slides
Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Maren Bennewitz,

Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Maren Bennewitz,

Introduction to Mobile Robotics SLAM Grid-based FastSLAM Wolfram Burgard, Maren Bennewitz, Diego Tipaldi, Luciano Spinello 1 The SLAM Problem SLAM stands for simultaneous localization and mapping The task of building a map while

910 views • 41 slides
Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard, Cyrill Stachniss, Maren Bennewitz, Giorgio Grisetti, Kai Arras Slides by Kai Arras and Wolfram Burgard Last update: June 2010 The SLAM Problem SLAM

1.16k views • 50 slides
Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard,

Introduction to Mobile Robotics SLAM: Simultaneous Localization and Mapping Wolfram Burgard, Cyrill Stachniss, Maren Bennewitz, Kai Arras Slides by Kai Arras and Wolfram Burgard Last update: June 2010 The SLAM Problem SLAM is the process by

1.04k views • 54 slides
Compact, high-power superconducting electron linacs as irradiators for materials and radiation

Compact, high-power superconducting electron linacs as irradiators for materials and radiation

FERMILAB-SLIDES-19-055-DI Compact, high-power superconducting electron linacs as irradiators for materials and radiation processing Illinois Accelerator Research Center (IARC), Fermilab This manuscript has been authored by Fermi Research

594 views • 38 slides
Q4 and Full Year 2018 Financial Results April 2, 2019 Tracy Pagliara Tim Howsman President and

Q4 and Full Year 2018 Financial Results April 2, 2019 Tracy Pagliara Tim Howsman President and

Q4 and Full Year 2018 Financial Results April 2, 2019 Tracy Pagliara Tim Howsman President and CEO Chief Financial Officer OTCQX: WLMS Cautionary Notes * Note: Unless otherwise noted, all discussion is based upon continuing operations.

560 views • 20 slides
presentation 19 August 2010 Discussion topics Business update Tom Gorman, CEO FY10 results

presentation 19 August 2010 Discussion topics Business update Tom Gorman, CEO FY10 results

2010 full-year results presentation 19 August 2010 Discussion topics Business update Tom Gorman, CEO FY10 results analysis Greg Hayes, CFO Outlook Looking forward growth Tom Gorman, CEO opportunities 2 Business update Tom Gorman,

980 views • 65 slides
Implementation PRESENTATION TO CIGIE AND GAOS FINANCIAL STATEMENT AUDIT CONFERENCE APRIL 28,

Implementation PRESENTATION TO CIGIE AND GAOS FINANCIAL STATEMENT AUDIT CONFERENCE APRIL 28,

Data Act Implementation PRESENTATION TO CIGIE AND GAOS FINANCIAL STATEMENT AUDIT CONFERENCE APRIL 28, 2015 1 DATA Act Summary Purpose: to establish government-wide financial data standards and increase the availability, accuracy, and

429 views • 14 slides
Development of a Multiphase Adjoint for CMAQ CMAS October 11, 2010 The team Carleton

Development of a Multiphase Adjoint for CMAQ CMAS October 11, 2010 The team Carleton

Development of a Multiphase Adjoint for CMAQ CMAS October 11, 2010 The team Carleton University (CU): ShunLiu Zhao and Amir Hakami Georgia Tech (GT): Shannon Capps, Athanasios Nenes, and Ted Russell University of Colorado

472 views • 24 slides
Project: More Experiments on Stochastic Gradient Methods Last updated: May 25, 2020 May 25, 2020

Project: More Experiments on Stochastic Gradient Methods Last updated: May 25, 2020 May 25, 2020

Project: More Experiments on Stochastic Gradient Methods Last updated: May 25, 2020 May 25, 2020 1 / 17 Goal We want to know more the internal details of simpleNN We want to roughly compare the two stochastic gradient approaches: SG with

502 views • 17 slides
The Prehistory and History of RE (+ SE) as Seen by Me: How My Interest in FMs Helped to Move Me

The Prehistory and History of RE (+ SE) as Seen by Me: How My Interest in FMs Helped to Move Me

The Prehistory and History of RE (+ SE) as Seen by Me: How My Interest in FMs Helped to Move Me to RE Daniel M. Berry University of Waterloo, Canada dberry@uwaterloo.ca See Slides 39--41, 46--51, and 60--65. They discuss how much of the

1.25k views • 80 slides
Learning for Children Birth Through Age Five: Effective Practices that Improve Outcomes Linda

Learning for Children Birth Through Age Five: Effective Practices that Improve Outcomes Linda

The Science of Dual Language Learning for Children Birth Through Age Five: Effective Practices that Improve Outcomes Linda M. Espinosa, Ph.D. Early Childhood Investigations Webinar May 11, 2016 Who is a Dual Language Learner? Terminology

826 views • 56 slides

More recommend