Learning Dynamic Manipulation Skills under Unknown Dynamics with - - PowerPoint PPT Presentation

▶
learning dynamic manipulation skills under
SMART_READER_LITE
LIVE PREVIEW

Learning Dynamic Manipulation Skills under Unknown Dynamics with - - PowerPoint PPT Presentation

Sep 02, 2023 25 likes •211 views

Learning Dynamic Manipulation Skills under Unknown Dynamics with Guided Policy Search Sergey Levine Pieter Abbeel UC Berkeley UC Berkeley Team TROOPER: Lockheed Martin, University of Pennsylvania, Philipp Krahenbuhl, Stanford University

slide-1
SLIDE 1

Learning Dynamic Manipulation Skills under Unknown Dynamics with Guided Policy Search

Sergey Levine

UC Berkeley

Pieter Abbeel

UC Berkeley

slide-2
SLIDE 2

Team TROOPER: Lockheed Martin, University of Pennsylvania, Rensselaer Polytechnic Institute Philipp Krahenbuhl, Stanford University

slide-3
SLIDE 3

general-purpose neural network controller +

slide-4
SLIDE 4

supervised learning trajectory optimization

policy search (RL) supervised learning trajectory optimization complex dynamics complex policy complex dynamics complex policy complex dynamics complex policy HARD EASY EASY

slide-5
SLIDE 5

prob

guided policy search

Trajectory Optimization

slide-6
SLIDE 6

approximate solution using iterative LQR (similar to extended Kalman filter)

  • locally linear dynamics
  • locally quadratic cost
  • Gaussian distribution

Trajectory Optimization

slide-7
SLIDE 7

Trajectory Optimization

slide-8
SLIDE 8

new

  • ld

Trajectory Optimization

slide-9
SLIDE 9
slide-10
SLIDE 10

Trajectory Optimization

slide-11
SLIDE 11
slide-12
SLIDE 12

Trajectory Optimization

slide-13
SLIDE 13

Trajectory Optimization

slide-14
SLIDE 14

Guided Policy Search

see Levine & Koltun, ICML 2014

slide-15
SLIDE 15
slide-16
SLIDE 16
slide-17
SLIDE 17
slide-18
SLIDE 18

Concluding Comments

  • simple linear dynamics model
  • fast, simple, standard LQR solver
  • can handle contacts despite linear

model

  • fit very complex policies with guided

policy search