Deep Learning Helicopter Dynamics Models
Ali Punjani Pieter Abbeel UC Berkeley EECS
Deep Learning Helicopter Dynamics Models Ali Punjani Pieter Abbeel - - PowerPoint PPT Presentation
Deep Learning Helicopter Dynamics Models Ali Punjani Pieter Abbeel UC Berkeley EECS Latent State: Airflow, Flexibility, Engine Dynamics etc. Similar trajectories have similar dynamics acceler acceleration ation stat state-contr e-control
Ali Punjani Pieter Abbeel UC Berkeley EECS
acceler acceleration ation stat state-contr e-control tr
ajector
acceler acceleration ation stat state-contr e-control tr
ajector
2 4 6 8 10
time (s)
−20 −10 10 20 30 40
Up-Down Acc. (ms−2)
circles
Observed Linear Acceleration Model ReLU Network Model
2 4 6 8 10
time (s)
−40 −30 −20 −10 10 20
Up-Down Acc. (ms−2)
flips-loops
Observed Linear Acceleration Model ReLU Network Model
2 4 6 8 10
time (s)
−40 −30 −20 −10 10 20 30 40
Up-Down Acc. (ms−2)
freestyle-aggressive
Observed Linear Acceleration Model ReLU Network Model
Ground Truth Baseline Model Our Model
2 4 6 8 10 RMS up-down acceleration error (ms−2)
forward-sideways-flight
vertical-sweeps stop-and-go turn-demos1 inverted-vertical-sweeps
dodging-demos4 circles flips-loops chaos turn-demos2 dodging-demos3 tictocs dodging-demos1 dodging-demos2 freestyle-gentle freestyle-aggressive turn-demos3
Up-Down Acceleration Error
Linear Acceleration Model Linear Lag Model Quadratic Lag Model ReLU Network Model