Skip to content

🤖 Robot Learning

EmbodiedGen-generated worlds are not just viewable — they are online training environments. Spin up parallel gym environments from a generated layout, record sensor and trajectory data, and evaluate the grasp quality of generated assets.

In a companion sim-to-real RL study, policies trained purely in EmbodiedGen-generated worlds reached 79.8% simulation task success (from 9.7%) and 75.0% real-robot task success (from 21.7%). The training code is not part of this repository; the tools below provide the environment side.

🏎️ Parallel Simulation Environments

Generate multiple parallel simulation environments with gym.make and record sensor and trajectory data.

python embodied_gen/scripts/parallel_sim.py \
--layout_file "outputs/layouts_gen/task_0000/layout.json" \
--output_dir "outputs/parallel_sim/task_0000" \
--num_envs 16
parallel_sim1 parallel_sim2

The generated results are organized as follows:

outputs/parallel_sim/task_0000
├── 0.mp4               # Tiled preview render across the parallel envs
├── layout.json         # Copy of the layout the envs were built from
└── scene_tree.jpg      # Scene-graph visualization

🦾 Grasp-Quality Evaluation

Evaluate the grasp quality of a generated URDF via a top-down Panda grasp trial (ManiSkill + SAPIEN). Outputs a JSON report and per-trial MP4 next to the URDF.

python embodied_gen/scripts/eval_collision_success.py \
  --urdf-path outputs/imageto3d/sample_00/result/sample_00.urdf \
  --num-trials 4

Next Steps