Propose
The frozen policy proposes a motion chunk. Its end-effector waypoints make the planned movement inspectable.
→ proposed actionsThe policy places the 8 and 7. GPT-6 helps recover the grasp and straighten the 2, handing control back between corrections.
Same task. Same layout.Videos show control-step playback at 25 fps. Model inference waiting time is omitted. Play each rollout using its own controls.
A little help. A more capable robot.
Selective intervention improves a frozen policy while leaving most actions to it.
Astra on Call: When and How GPT-6 Should Intervene in a VLA Policy
Better assistance means recovering mistakes while preserving useful behavior. Selective takeover does both on the evaluated tasks.
Weak-set Score · 35 episodes
Policy only → selective takeover
Aggregate Score · 15 episodes
Policy only → selective takeover
Weak set · versus full GPT-6 control
Using the paper’s timing accounting
7 tasks · 35 episodes
Score = mean partial credit, from 0 to 100. It is not success rate.
Timing: 1,269 s vs 1,753 s per weak-set episode, accounting for 16 s per model call, 0.3 s per control step, and 3 s per consultation. These figures describe the study’s execution setup.
The biggest advantage over full GPT-6 control comes from keeping more of the policy’s original successes.
Fixed-15 across the matched weak and strong sets. Full GPT-6 control preserves 2/13 and recovers 7/37.
A scene, a proposed motion, and a choice. The model inspects what the policy intends to do before deciding whether to help. No policy retraining.
The frozen policy proposes a motion chunk. Its end-effector waypoints make the planned movement inspectable.
→ proposed actionsGPT-6 considers the scene, task, execution feedback, and proposed motion. Fixed-15 consults every 15 policy steps.
Scene + plan + feedbackKeep the policy in control, take over an action segment, or—in the editing variants—adjust the proposed waypoints.
Then observe and repeat ↺Bounded waypoint editing lowers weak-set Score at three of four evaluated Fixed-K intervals. At K = 15, segment takeover scores 42.9, compared with 25.1 for EditOnly.
The correction interface matters: a successful local adjustment in one rollout does not imply a better controller overall.
3 of 4 intervals favor segment takeover.RoboDojo manipulation tasks probe both recovery and preservation: seven tasks where the baseline struggles, and three where it already performs well.
These are results from simulated manipulation with a frozen policy. Fixed-15 is an observed strong setting on this evaluation, and aggregate preservation does not mean that every individual success is retained.
Follow the robot through recovery, unsuccessful endings, and different correction interfaces. Each figure preserves the original frames, decision timeline, and legend.