adapting-pretrained-vla-manipulation-policy-only-handful-target-demonstrations CLM-0543mechanismsingle paperReviewed by AI
When adapting a pretrained VLA manipulation policy with only a handful of target demonstrations, retrieving prior demonstrations at the level of LLM-decomposed subtasks (then reranking by behavior-feature similarity) raises success more than retrieval by visual, optical-flow, or state-action similarity, and the margin is concentrated on long-horizon composite tasks rather than single-skill pick-and-place.
Capability: Turning plans into physical action
Observed on
SmolVLA-450M backbone; 5 demos per LIBERO long-horizon task with LIBERO-90 as prior, 20 demos per real xArm task with 10k DROID trajectories as prior; manipulation only.
Sources
- Validated by: Average success rate over 50 episodes x 3 seeds on 10 LIBERO long-horizon tasks (43.5% vs 33.2% for Behavior Retrieval, 31.9% STRAP, 32.5% IWR, 22.7% BC) and 20 trials per task on four real xArm tasks (48/80 vs 31/80 IWR, 27/80 BC); ablations remove decomposition, reranking, and the two-stage split.. Measured on one benchmark (LIBERO long-horizon, 10 tasks, 3 seeds) and 4 real-robot tasks with 20 trials each, against six baselines including BC, random, all-data, and four retrieval methods. Single VLA backbone (SmolVLA-450M), single LLM planner. [truncated]
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- On language-conditioned manipulation tasks where the same base task is instantiated at increasing levels of spatial ambiguity and procedural length, current VLA models degrade on fine-grained spatial relations, precise low-level execution, and memory-dependent multi-step planning, so success measured only on predefined simple settings overstates their competence.Turning plans into physical action · reviewed by AI
- For post-training a pretrained vision-language-action manipulation policy under a fixed per-round frame budget, collecting handheld demonstrations only at states where the human action chunk diverges from the policy's sampled action distribution, and weighting them by a learned task-progress advantage, raises task progress more than collecting the same number of frames of ordinary demonstrations and fine-tuning on them.Turning plans into physical action · reviewed by AI
- For manipulation policies whose large VLA backbone cannot run within the control period, letting the slow model propose action chunks from a stale observation while a small RL-trained edit policy adjusts and Q-selects among those chunks using the latest observation recovers more performance than applying RL on top of asynchronous imitation-learning schemes alone, and the gap widens as inference delay or environment speed grows.Turning plans into physical action · reviewed by AI
- When a pretrained vision-language-action policy is fine-tuned sequentially on a stream of real-robot manipulation tasks, the choice of action-normalization statistics is itself a determinant of continual-learning success: statistics that are task-derived, first-task-derived, or recomputed each stage produce forgetting or poor transfer, while statistics estimated once from a small task-independent calibration set covering the embodiment's motion range and then frozen give the best acquisition, retention and task-order stability.Turning plans into physical action · reviewed by AI
- For agentic retrieval-augmented multi-hop QA with 7B-class policy models, scoring each intermediate step for both logical validity and grounding in the retrieved passages — rather than scoring steps by the final answer — yields higher EM/F1 than outcome-reward and final-answer-based process-reward training, because it removes preference pairs where flawed retrieval happened to reach the right answer.Carrying a chain of inference through to its conclusion · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A matched-budget comparison where skill-level language retrieval plus reranking does no better than state-action or subtrajectory retrieval on long-horizon composite tasks, or where its advantage is as large on single-skill tasks as on composite ones.