ALLSTEPS: Curriculum-driven Learning of Stepping Stone skills @m.vandepanne

Michiel van de Panne | ALLSTEPS: Curriculum-driven Learning of Stepping Stone skills @m.vandepanne | Uploaded May 2020 | Updated October 2024, 2 hours ago.
This video accompanies the paper of the same title.
Curriculum-driven deep reinforcement learning is used to learn control policies for physics-based characters, enabling them to walk across stepping stone sequences and variable terrain. The control policy perceives the upcoming two desired stepping locations.

SCA 2020: ALLSTEPS Curriculum-driven Learning of Stepping Stone Skills (full talk)

Dynamic Terrain Traversal Skills Using Reinforcement Learning (part 1)

Learning to Locomote: Understanding How Environment Design Matters for Deep Reinforcement Learning

Feedback Control for Cassie with Deep Reinforcement Learning

Progressive Reinforcement Learning with Distillation for Multi-Skilled Motion Control

Dynamic Animation Synthesis with Free-Form Deformations

PartwiseMPC: Interactive Control of Contact-Guided Motions

SIGGRAPH 2017 DeepLoco (supplemental video)