All pull requests

Pull requests

• mixer_operating: enable the farthest-candidate fallback for distractor sampling

  #89·liyifreddy opened Oct 12, 2026Contributor·

  ·
  1

• smolvla: check success after every step inside the action chunk

  #87·liyifreddy opened Oct 12, 2026Contributor·

  ·
  1

• smolvla: seed the inference worker from the episode seed

  #85·liyifreddy opened Oct 12, 2026Contributor·

  ·
  1

• Keep the farthest candidate when distractor sampling runs out of attempts (drawer_open_place)

  #83·liyifreddy opened Oct 12, 2026Contributor·

  ·
  1

• fix: resolve SmolVLA eval silent failures (#76)

  #77·Jah-yee opened Oct 3, 2026·

  ·
  1

• configs: record drawer_pose / pipette_pose with get_object_pose

  #74·liyifreddy opened Sep 28, 2026Contributor·

  ·
  1

• items_handover: compute the pen/holder height overlap per env

  #72·liyifreddy opened Sep 28, 2026Contributor·

  ·
  1

  1

• Save ACT checkpoints on rank 0 only in distributed training

  #70·liyifreddy opened Sep 28, 2026Contributor·

  ·
  1

• ACT train.py: call lerobot's init_logging() so the loss is printed

  #66·liyifreddy opened Sep 27, 2026Contributor·

  ·
  1

• Comment why compute_task_state() returns zero success/fail

  #61·liyifreddy opened Sep 27, 2026Contributor·

  ·
  1

• Update dexsim-engine 0.4.3 hashes in policy/act and policy/dp uv.lock

  #58·liyifreddy opened Sep 25, 2026Contributor·

  ·
  1

• SmolVLA adapter: pass float32 images in [0, 1] and return four values from eval()

  #57·liyifreddy opened Sep 25, 2026Contributor·

  ·
  1

• feat: expose trajectory recording and harden task launches

  #36·yuecideng opened Aug 6, 2026Collaborator·13 of 15 tasks·

  ·

• feat: add Viser browser visualization support

  #33·yuecideng opened Jul 30, 2026Collaborator·11 tasks done·

  ·

• Add atomic replay migration for RoboSyn action banks

  #15·skywhite1024 opened Jun 25, 2026Collaborator·

  ·