VLAarchtests3 / setup /ENVIRONMENT.md
lsnu's picture
Add files using upload-large-folder tool
aa584de verified

Environment Manifest

This export was assembled on:

  • OS: Ubuntu 22.04.5 LTS
  • Kernel: Linux 6.8.0-88-generic
  • GPU: NVIDIA L40S
  • VRAM: 46068 MiB
  • Driver: 580.126.09
  • Python: 3.10.20

Primary Python Environment

The main RLBench-capable environment used during this handoff lived at:

  • /workspace/envs/rlbench

The exact package snapshot from that environment is stored in:

  • setup/rlbench_pip_freeze.txt

Upstream Pins

Pinned benchmark stack used by this project:

  • peract_bimanual: bb0232a6ba3fe116566e9568f0c7af980ed6703d
  • RLBench: 8af748c51287989294e00c9c670e3330a0e35ed5
  • PyRep: b8bd1d7a3182adcd570d001649c0849047ebf197
  • YARR: 6822ff78602c77878b27d4cfe759ce029c67bffb
  • AnyBimanual: 76024e48b0e9489101459e85bc909c126ec581b4

Important Runtime Variables

The RLBench / Coppelia / AnyBimanual stack was run with environment variables equivalent to:

export DISPLAY=:99
export XDG_RUNTIME_DIR=/workspace/runtime
export COPPELIASIM_ROOT=/workspace/assets/coppeliasim_v4_1_0
export QT_QPA_PLATFORM_PLUGIN_PATH=/workspace/assets/coppeliasim_v4_1_0
export LD_LIBRARY_PATH=/workspace/assets/coppeliasim_v4_1_0:${LD_LIBRARY_PATH:-}
export PYTHONPATH=/workspace/third_party/PyRep:/workspace/third_party/RLBench:/workspace/third_party/YARR:/workspace/third_party/AnyBimanual

For the local project code, the handoff runs also used:

export PYTHONPATH=/workspace/workspace/VLAarchtests2_code/VLAarchtests/code/reveal_vla_bimanual:$PYTHONPATH

Notes

  • RLBench headless execution still required an X server.
  • The abstract reveal/retrieve proxy benchmark did not depend on RLBench or CoppeliaSim.
  • The official AnyBimanual dual_push_buttons path was the only general-task anchor treated as trustworthy on this setup.