Skip to content

Add mujoco_emdb_sim: RoboCasa/robosuite lift experiment - #7

Open
jpperezm wants to merge 3 commits into
latestfrom
feat/robocasa-lift
Open

jpperezm wants to merge 3 commits into
latestfrom
feat/robocasa-lift

Conversation

@jpperezm

Copy link
Copy Markdown

New mujoco_emdb_sim package bridging SantaCRC's RoboCasa/robosuite KitchenLift
sim to the e-MDB architecture:

  • sim_bridge: hosts executed_action/world_reset, forwards to scene_loader's
    /step_action + /reset_episode
  • EmdbSimulatorPerception: flattens ObjectStateArray + scalar mdb sensors
  • perception_relay: TEMPORARY keep-alive for the sim's one-shot rl-mode
    perceptions (remove once scene_loader streams them)

Also adds lift_experiment.yaml + lift_launch.py (PolicyLearned/SAC on the lift
task) and the ur5_grasp experiment config/launch.

Verified running end-to-end on latest: single clean architecture, perceptions
flow, world model learns, actions are selected.

jpperezm and others added 3 commits August 12, 2026 13:53
New package mujoco_emdb_sim bridging emdb_simulator (SantaCRC's RoboCasa/
robosuite KitchenLift sim) to the e-MDB architecture:
  - sim_bridge: hosts executed_action/world_reset, forwards to scene_loader's
    /step_action + /reset_episode
  - EmdbSimulatorPerception: flattens ObjectStateArray + scalar mdb sensors
  - perception_relay: TEMPORARY keep-alive for scene_loader's one-shot rl-mode
    perceptions (remove once the sim streams them)

Adds lift_experiment.yaml + lift_launch.py (PolicyLearned/SAC on the lift task)
and the ur5_grasp experiment config/launch.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- EmdbSimulatorPerception: pass base Perception args by keyword (latest inserted
  a node_type param, which broke the positional super().__init__ call).
- sim_bridge: ignore control-topic "reset_world" when the world_reset service is
  hosted, so two callbacks don't drive the shared reset ServiceClient concurrently
  ("generator already executing").

Verified end-to-end on latest: single clean architecture, perceptions flow,
world model learns, actions are selected.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants