Festivus

policy

DreamWaQ

KAIST · PyTorch

or hover any field below to flag it

Overview

Name
DreamWaQ
Author
KAIST
Framework
PyTorch
License
other
Skill type
locomotion
Evidence level
verified
Task description
RL-based quadrupedal locomotion using implicit terrain imagination from proprioceptive sensing only. Learns a latent terrain representation via teacher-student distillation, enabling gait adaptation without exteroceptive sensors. Robust outdoor sim-to-real transfer on challenging terrain.

Spaces

Action space
joint-position · 12-dim · 50Hz
Observation space
  • type: proprioception
  • · body_angular_vel (3)
  • · gravity_vector (3)
  • · velocity_command (3)
  • · joint_angles (12)
  • · joint_angular_vel (12)
  • · previous_action (12)

Links

HuggingFace repo
null
Paper (arXiv)
https://arxiv.org/abs/2301.10602

Compatible robots

0+1 mentioned but not in catalog yet

No robots list DreamWaQ as compatible yet. Know of one? Flag it above.

Compatible environments

2
flat-terrainnot in seedrough-terrain-outdoornot in seed

Datasets that reference this policy

0

No datasets reference DreamWaQ yet.