Event
ElevateHead: An Action Head that Decides Once and Commits
1 reports · 1 sources
Tsung-Lung Yang et al. released a paper proposing the ElevateHead action head: it selects the task execution method once at a learned decision point and keeps it fixed throughout the episode; users can also set this choice as an input. On held-out Franka Kitchen tasks, the proportion of episodes in which ElevateHead fully completed all four subtasks was 45.2%, versus 29.3% for BeT (+0.160, 95% CI [+0.129, +0.188]). However, the authors state that this trend did not reproduce on other tasks: on a held-out block stacking task, version 4 was comparable to but did not exceed BeT; each of the first four versions was below BeT on at least one held-out task. The authors state that a real-robot test plan using pre-fixed predictions will be presented in a subsequent version of the paper.
Report timeline
- ElevateHead: When Action Heads Need One-Time Decisions in Multimodal Imitation Learning
ElevateHead proposes an action head that selects the task execution mode once at a learned decision point and keeps it fixed for the entire episode; users can also set this choice as an input. On held-out Franka Kitchen tasks, 45.2% of its episodes completed all four subtasks, versus 29.3% for BeT (+0.160, 95% CI [+0.129, +0.188]); on held-out block stacking tasks, version 4 was comparable to BeT but did not surpass it.
arXiv cs.RO ·