SLIDE 1
Guided Policy Search Sergey Levine Learning on PR2 Shape sorting - - PowerPoint PPT Presentation
Guided Policy Search Sergey Levine Learning on PR2 Shape sorting - - PowerPoint PPT Presentation
Guided Policy Search Sergey Levine Learning on PR2 Shape sorting cube Visuomotor Policies Guided Policy Search trajectory optimization supervised learning expectation under current policy trajectory distribution(s) Lagrange multiplier
SLIDE 2
SLIDE 3
Shape sorting cube
SLIDE 4
Visuomotor Policies
SLIDE 5
Guided Policy Search
supervised learning trajectory optimization
SLIDE 6
expectation under current policy
trajectory distribution(s)
Lagrange multiplier
SLIDE 7
Supervised Learning Objective
SLIDE 8
Trajectory Optimization (without GPS)
SLIDE 9
Trajectory Optimization
SLIDE 10
new
- ld
Trajectory Optimization
[see Levine & Abbeel ‘14 for details]
SLIDE 11
[see L. et al. NIPS ‘14 for details]
SLIDE 12
Trajectory Optimization (with GPS)
SLIDE 13
[see L. et al. NIPS ‘14 for details]
SLIDE 14
Instrumented Training
training time test time
SLIDE 15
~ 92,000 parameters
Chelsea Finn
SLIDE 16
Experimental Tasks
SLIDE 17
Shape sorting cube
SLIDE 18
Hanger
SLIDE 19
Hammer
SLIDE 20
Bottle
SLIDE 21
Locomotion
Igor Mordatch
better trajectory optimization + large scale simulation
SLIDE 22
Darwin Robot
Igor Mordatch
better trajectory optimization + large scale simulation + adaptation to real world dynamics
Mordatch, Mishra, Eppner, Abbeel
SLIDE 23
Guided Policy Search Applications manipulation locomotion
with N. Wagener and P. Abbeel with V. Kumar and E. Todorov with V. Koltun
aerial vehicles
with G. Kahn, T. Zhang, P. Abbeel
tensegrity robot
with M. Zhang, K. Caluwaerts, P. Abbeel
dexterous hands
SLIDE 24
DAGGER
typically 0.0, except when i = 1, then 1.0
SLIDE 25
DAGGER Video
See http://videolectures.net/aistats2011_ross_reduction/
SLIDE 26
Trajectory Optimization – Dynamics Fitting
SLIDE 27
[see L. et al. NIPS ‘14 for details]
SLIDE 28
Learned Motion Skills
SLIDE 29
More Visuomotor Experiments
SLIDE 30
Beyond Instrumented Training
training time test time
Finn, Tan, Duan, Darrell, L., Abbeel ‘15
SLIDE 31
Learning Visual State Spaces
SLIDE 32