Robots improve task success by committing to one action plan early
ElevateHead: Decide Once, Then Commit: When the Action Head Matters for Multimodal Imitation
Robotics
Summary
Robots learning from human demonstrations can see multiple ways to do the same task, like going around an obstacle from different sides. The authors found that if a robot keeps deciding what to do at every step, it may switch back and forth and get confused. They developed a method called ElevateHead that makes the robot pick one way early on and sticks with it throughout the task, improving success rates. This approach also lets a user choose the way the robot will take. Tests showed their method outperformed a strong baseline in a kitchen task and identified where sticking to a plan breaks down.
What this means in practice
- •For robotics engineers: Improve robot task completion by implementing early commitment to one decision mode based on ElevateHead, reducing indecision during execution.
- •For automation system integrators: Integrate user-settable decision modes in robots to control task pathways more reliably in environments with multiple valid strategies.
Authors
Tsung-Lung Yang
Abstract
Robots that learn from people often see several valid ways to do the same task: pass an obstacle on the left or on the right, open the microwave or move the kettle first. A policy that re-decides at every step can waver between these ways mid-task. We ask when the action head (the part of a policy that turns the state into actions) matters, and how it should make this choice. On a controlled task, heads that sample one way succeed far more often than a head that averages the ways when the demonstrations contain several ways and the state gives no hint which to take (+0.72 success with two ways); this trend was not reproduced on the other tasks we ran. We then present ElevateHead, a head that decides once at a learned decision point, commits to that choice for the episode, and exposes the choice as an input a user can set. We designed it by diagnosis: each version was frozen and tested on tasks it had not seen, and the next version changed only what a measured failure pointed to; predictions were registered before each run. The first four versions each fell below BeT on at least one held-out task. The fifth completed all four subtasks of the held-out Franka Kitchen task in 45.2% of episodes against 29.3% for BeT, the strongest baseline across our tasks (+0.160, 95% CI [+0.129, +0.188]); when a mode was forced, the first completed subtask matched it in 95% of successful episodes. Diagnosis also located where commitment breaks: varied start positions (repaired on the controlled task: commitment 0.22 to 0.81, success 0.74 to 0.91) and varied object placement (partly repaired: in an exploratory variant, commitment 0.03 to 0.59 on aligning with no detectable change in success). On held-out block stacking, version 4 was comparable to BeT and did not exceed it. A real-robot test with predictions fixed in advance is planned for a follow-up version of this paper.