Does ACT run on WidowX 250 S?

Capabilities

Caveats

• The BridgeData V2 evaluation used goal-image-conditioned ACT. bridge_data_v2 does not include the ACT implementation or a checkpoint.

Evidence 3 sources; the cell shows the highest tier

1. T3 · community reportedhub-checkpointsource 2025-10-12verified 2026-10-24 (1d ago)

   config.json type=act. train_config dataset msnaga/widowx_250_pick_and_place_block, whose meta/info.json robot_type=widowx_250s_end_effector (50 episodes, one front camera, waist/shoulder/elbow/forearm_roll/wrist_angle/wrist_rotate/gripper joints). No model card or rollout results.

   https://huggingface.co/msnaga/widowx_250_pick_and_place_block_policy
2. T3 · community reportedpapersource 2024-01-17verified 2026-10-24 (1d ago)

   BridgeData V2 paper (dataset paper; ACT first author Tony Zhao is a co-author) trains goal-conditioned ACT on BridgeData V2 and evaluates it on the real WidowX 250, 10 trials per task (Table 2).

   “Action chunking with transformers (ACT). Previous work has also noted that predicting action sequences instead of single actions can benefit imitation learning [17, 18].”

   https://arxiv.org/abs/2308.12952v3
3. T1 · upstream documentedsource-codesource 2024-03-17verified 2026-10-24 (1d ago)

   bridge_data_v2 implements GCBC, D-GCBC, GCIQL, CRL and LCBC only. ACT is linked to its upstream repo, and no ACT checkpoint is released.

   “We don't currently have a checkpoints for ACT or RT-1 available but may release them soon.”

   https://github.com/rail-berkeley/bridge_data_v2/blob/bc60a35b701a12021c8c95e9d8601274d3acd928/README.md