Navigation Around Humans Hey!! How you Do' in Importance, - - PowerPoint PPT Presentation

Navigation Around Humans

Importance, Approaches and the Future!!!

Hey!! How you Do' in

WHY IS IT IMPORTANT??

That’s Why

Papers Presented and Discussed

• 1. Adaptive Human aware Navigation

based on Motion Pattern Analysis

(Søren Tranberg Hansen, Mikael Svenstrup, Hans Jørgen Andersen and Thomas Bak)

• 2. Socially-Aware Robot Navigation: A

Learning Approach

(Matthias Luber, Luciano Spinello, Jens Silva, Kai

• O. Arras)

Do It Like a BOSS!!!

Paper 1 Abstract

• —Respecting people’s social spaces is an important prerequisite for

acceptable and natural robot navigation in human environments.

• In this paper, we describe an adaptive system for mobile robot

navigation based on estimates of whether a person seeks to interact with the robot or not. The estimates are based on run- time motion pattern analysis compared to stored experience in a database.

• Using a potential field centered around the person, the robot

positions itself at the most appropriate place relative to the person and the interaction status. The system is validated through qualitative tests in a real world setting.

Peek-a-boo

Components of the Paper

• 1. Introduction
• 2. Material and Methods
• 3. Experimental Setup
• 4. Results
• 5. Discussion
• 6. Conclusion

Introduction

• Need to develop techniques that determine whether the user wants

to interact with the robot or not in that current situation(example: robots supporting care assistants)

• Evaluating and Predicting whether the user wants to interact with

the robot based on the position and pose of the user.

• Use the CBR(Case Based Reasoning) in real world environment

scenario to determine whether the user would like to interact or not

• CBR allows recalling and interpreting past experiences, as well as

generating new cases to represent knowledge from new experiences

Umm… Hi!

Materials and Methods

• Divide the Zone around People into four zones:
• 1. the public zone > 3.6m
• 2. the social zone > 1.2m
• 3. the personal zone > 0.45m
• 4. the intimate zone < 0.45m
• If likely that the person does not seek to interact, then do not

violate the personal zone and stay in the public or the social zone.

• If likely that the person does seek to interact, then try to move

into the personal zone.

I told you!! she was not interested.

Materials and Methods(Continued)

• Evaluator used to predict this behaviour.
• Philosophy of the Evaluator:
• a. human motion pattern is relative to the robot and the

chances of an interaction

• b. Can be estimated based on the pose and position of the

human and the corresponding stored interaction information for the same from the previous data.

• Use this data to control the robot and set its objectives

accordingly.

YO bro!! You thinking? No bro!! Evaluating

Materials and Methods(Still going on)

• Evaluating Human robot encounters:
• 1. Variables:
• a. PI: fuzzy logic used to determine chances of interaction.

(1=close interaction; 0=no close Interaction)

• b. X,Y: 2D coordinates of the human’s position in respect to the

robot being the origin for the plane.

• c. Theta: angle of pose in respect to robot.
• d. W: weight assigned to PI according to proximity to the robot
• f the human(closer proximity to robot, more the weight).

Materials And Methods

• Two Stages of Robot Interaction:
• 1. A person encounters the robot, and the robot evaluates the

person given all the previous experiences from the database.

• 2. the robot updates the database according to the person

encounter, which has just passed.

Materials and methods(Conclusion)

• Algorithm I
• if (Interested) then PI = PI + wL
• if PI > 1 then PI = 1
• else if (Not Interested) then PI = PI - wL
• if PI < 0 then PI = 0
• For modeling the robots navigation system, a person centered

potential field is introduced.

• The potential field is calculated by the weighted sum of four

Gaussian distributions of which one is negated. The covariance of the distributions are used to adapt the potential field according to PI.

ZZZZZZZZ

Materials And methods(Conclusion continued)

• robot will navigate in such a way to reach the dark blue region

and stay in it( Steepest Descent Approach).

• The potential field has varying color zones according to the

values of PI.

Experimental Setup

• Evaluation of method performed through two experiments.
• Experiment 1:

a. the objective was to see if estimation of PI can be obtained based on interaction experience from different persons. b. A total of five test persons were asked to approach or pass the robot using different motion patterns c. The starting and end point of each trajectory were selected randomly, while the specific route was left to the own devices of the test person d. Random selection designed such that the cases with interaction were 50% and cases with no interaction also 50%. e. The output values (PI), the input values (position and pose), and the database were logged for later analysis.

Experimental Setup

• Experiment 2
• a. Objective: test the adaptiveness of the method. The system should be

able to change its estimation of PI over time for related behavior patterns.

• b. Total of 36 test approaches performed with one test person.

c. The test person would start randomly from any three random positions and end his trajectory in fixed position.

• d. First 18 encounters, test person shows interest in interaction. Last 18

encounters, test person does not show interest.

• e. The output values (PI), and the input values (position and pose) were

logged for later analysis.

He he he… Perfect alignment

Experimental Setup

Results

• Experiment 1
• a. The starting point of the CBR is an empty database. As the robot-

human encounters take place, the database gets gradually filled up.

• b. We use 5 test persons for the database development and

evaluation.

• c. We use 4D plots to display the results of our experiments.
• d. 2 dimensions for 2D position of the human with respect to the

robot.

• e. 1 dimension to portray the direction or pose of the human.
• f. 1 dimension to depict the probability of interaction.

Results(Continued)

Results(conclusion)

• Experiment 2
• a. Objective of the experiment was to show that estimation of

PI will adapt based on our observations.

• b. The values of PI in data are averages of PI for three areas.

Discussion

• What Did we like about the Paper??
• What We didn’t Like??
• Where can we employ this approach??
• Any other Suggestions for the paper.

Paper 2 Abstract

• The ability to act in a socially-aware way is a key skill for robots that

share a space with humans.

• In this paper we address the problem of socially-aware navigation

among people that meets objective criteria such as travel time or path length as well as subjective criteria such as social comfort.

• Opposed to model based approaches typically taken in related work,

we pose the problem as an unsupervised learning problem.

• We learn a set of dynamic motion prototypes from observations of

relative motion behavior of humans found in publicly available surveillance data sets.

• The learned motion prototypes are then used to compute dynamic cost

maps for path planning using an any-angle A* algorithm.

Introduction

• Research in the area of socially-aware navigation and manipulation is

typically taking a model based approach, either with manually designed models or models from social psychology and cognitive science

• However, there is a methodological gap as all these models have been

tested or evaluated in controlled environments.

• The people who wrote the paper believe, that the socially aware

behaviour of the system should be learned from real world data.

• the authors address the problem of learning a planning strategy

through streams of pedestrians rather than learning continuous sensory- motor motions.

• we take annotated surveillance data sets collected from overhead

cameras, extract the pedestrian paths, and transform them into a 3D representation

Components of the Paper

• Learning Relative Motion Prototypes
• Planning with RMPs
• Experiments
• Conclusion

Learning Relative Motion Prototypes

• In this section we present the theory for learning socially-aware relative

motion prototypes (RMP).

• Because of various complexities involved in handling situations with

multiple humans, we break down the problem into pairwise evaluations.

• A relative motion prototype Ri,j describes a relative motion between

person πi and person πj.

• Given the two observation sequences zi and zj of their (x,y)-positions
• ver time t, we define
• di,j(t) = MOD (kzi(t)−zj(t)k)
• Ri,j = [di,j(ts), ... ,di,j(te)]
• Ri,j = {di,j(ts),...,di,j(te)}

Learning Relative Motion Prototypes

• Next step is to cluster the results.
• As relative movement sequences can difger in duration and

relative speed, we have to define an appropriate distance function able to group similar motion behaviors into the same cluster

• Cluster=Prototype
• ADTW technique employed for the same
• Extension of DTW
• Used for matching sequences of words spoken at different speeds.

Learning Relative Motion Prototypes

Learning Relative Motion Prototypes

• Results after first step are clusters that are meaningless as the

goal and other variables that influence human motion are not considered yet.

• One aspect that we use to distinguish between the various

clusters is the angle of approach criterion.

Learning Relative Motion Prototypes

• We determine the optimal number n of social contexts, that is

the optimal number of αi,j-intervals.

• We finally define P = {Rα1 i,j,....,Rαn i,j}

to be the set of relative motion prototypes composed of n elements that refer to n social contexts

Planning with RMP’s

• This section presents the methods employed to derive socially-aware paths from the learned

models.

• Done in two phases:
• a. Social Context Detection

1. Deals with finding the right prototype for that situation. 2. The Selected Prototype is time Warped to match the velocity of the approaching subject.

• b. Cost Map Computation and Path Planning
• 1. The selected motion prototype is then used to derive a dynamic cost map in a regular grid

tessellation of the environment.

• 2. This is done by computing a time varying cost function for each grid cell that follows the

learned distances di,j(t) as specified by the prototype.

• 3. As cost function we use a Gaussian distribution with mean z(t) and standard deviation σ =

di,j(t)/3 at time t laid over the grid.

• 4. The costs are updated in each step over the entire duration of the interaction

Planning with RMP’s

Experiments

• A. Data set
• B. Training
• C. Qualitative Experiments

Experiments Are Always important

Experiments

• Data Set:
• a. We used EIPD database to learn the behaviour of the walking

people.

• Training:
• a. We have taken 90 pairwise interactions of people consisting in

180 tracks to learn the relative motion

• b. Special treatment taken for still person case as that case is

undefined.

• c. Observed some limitations to the proxemics approach that

further solidifies this paper’s hypotheses.

Experiments

• Quantitative Results

Experiments

• Quantitative Results:

Whose brain is this??

Discussions

• Did you Like the Paper??
• Why Didn’t you Like the Paper??
• Was their Approach satisfactory??
• Were their Results believable??
• Any Suggestions??

So MANY QUESTIONS System Overload!!!

Thank YOU!!


BYEE!!!!