Issues

• mixer_operating: the distractor min distance (1.5 m) can never be satisfied inside the sampling box

  EDEM-AI/RoboSynChallenge#88 · Filter by author liyifreddy opened on Oct 12, 2026
  1

• SmolVLA adapter only checks success at the end of each action chunk, so successful episodes are charged extra steps

  EDEM-AI/RoboSynChallenge#86 · Filter by author liyifreddy opened on Oct 12, 2026
  1

• SmolVLA: the inference worker never sets a seed, so the same evaluation episode is not reproducible

  EDEM-AI/RoboSynChallenge#84 · Filter by author liyifreddy opened on Oct 12, 2026
  1

• drawer_open_place: a distractor (fork or spoon) sometimes spawns on/inside the cabinet and the drawer then can't be opened

  EDEM-AI/RoboSynChallenge#82 · Filter by author liyifreddy opened on Oct 12, 2026
  1

• SmolVLA eval fails silently without #53's errors ever surfacing: worker stdout buffering + tokenizer path bug (+ one rules question)

  EDEM-AI/RoboSynChallenge#76 · Filter by author Beeny0814 opened on Oct 3, 2026
  4

• Same seed, same machine: the scene after reset differs by a few mm between runs (item_assembly, drawer_open_place, sample_loading)

  EDEM-AI/RoboSynChallenge#75 · Filter by author liyifreddy opened on Sep 28, 2026

• drawer_pose and pipette_pose are all zeros (get_rigid_object_pose is used for articulations)

  EDEM-AI/RoboSynChallenge#73 · Filter by author liyifreddy opened on Sep 28, 2026

• items_handover: pen/holder height-overlap check uses only env 0 and is broadcast to all envs

  EDEM-AI/RoboSynChallenge#71 · Filter by author liyifreddy opened on Sep 28, 2026

• Distributed ACT training writes the same checkpoint once per rank

  EDEM-AI/RoboSynChallenge#69 · Filter by author liyifreddy opened on Sep 28, 2026

• policy/act/scripts/train.py never prints the training loss (init_logging() is not called)

  EDEM-AI/RoboSynChallenge#65 · Filter by author liyifreddy opened on Sep 27, 2026

• random setting: the cam_high extrinsics randomization has no effect (pos_range/euler_range are ignored for look_at cameras)

  EDEM-AI/RoboSynChallenge#62 · Filter by author liyifreddy opened on Sep 27, 2026
  1

• SmolVLA_sim_drawer_open_place and SmolVLA_sim_manipulate_pipette: the first 64 KiB of one frozen vision weight is all zeros

  EDEM-AI/RoboSynChallenge#59 · Filter by author liyifreddy opened on Sep 27, 2026

• uv sync --frozen in policy/act fails: hash mismatch for dexsim-engine 0.4.3

  EDEM-AI/RoboSynChallenge#55 · Filter by author liyifreddy opened on Sep 25, 2026

• SmolVLA adapter doesn't run under eval_policy.py (eval() returns 3 values, images passed as uint8)

  EDEM-AI/RoboSynChallenge#53 · Filter by author liyifreddy opened on Sep 25, 2026

• SmolVLA setup_lerobot.sh installs unpinned LeRobot main: released v2.1 datasets can't be loaded and finetune.sh fails

  EDEM-AI/RoboSynChallenge#52 · Filter by author liyifreddy opened on Sep 25, 2026