The paper investigates how to best utilize demonstration quality during vision‑language‑action (VLA) pretraining to boost downstream task performance. It compares weighting, filtering, and conditioning strategies across different reward definitions and label granularities, using 5,000 suboptimal human demonstrations and 24 target tasks while keeping demonstration sets and training budgets matched. The study introduces AXIS‑Bench, a benchmark built on crowdsourced simulation demonstrations that allows recomputation of quality labels and replayable trajectories for both simulation and real‑world tasks, showing that simulation pretraining can raise real‑world success rates from 9.6% to 31.7%.
arXiv:2606. 15631v1 Announce Type: cross Abstract: Extending a vision-language-action (VLA) policy to a new task typically requires task-specific teleoperated demonstrations and per-task fine-tuning, making adaptation costly in both data collection and compute.
arXiv:2609.13851v1 Announce Type: cross Abstract: Post-training vision-language-action (VLA) models for specific robots and tasks requires in-domain demonstrations, yet collecting diverse robot data...
The paper introduces AVA, a training‑free framework that improves interleaved multimodal in‑context learning for frozen large vision‑language models. AVA decomposes the interleaved context into per‑sample threads, selects influential demonstrations while removing redundancy, and calibrates consistency using a weak sibling model’s confidence. Across five vision‑language benchmarks, AVA raises performance by up to 2.4 points, narrows the visual performance gap on ScienceQA‑IMG, and remains robust even with 30% noisy demonstrations, all while keeping the LVLM frozen and adding only modest latency.
Vision‑Language‑Action models (VLAs) often fail to follow language instructions because they rely on vision shortcuts learned from biased datasets. The authors introduce LIBERO‑CF, a counterfactual benchmark that tests language following by providing alternative task instructions for the same scene layouts, revealing widespread counterfactual failures. They propose Counterfactual Action Guidance (CAG), a dual‑branch inference scheme that amplifies language conditioning, which consistently improves language grounding and task success across diverse VLAs without requiring extra data or architectural changes.
arXiv:2511. 16107v3 Announce Type: replace-cross Abstract: Visual in-context learning (VICL) solves visual tasks by conditioning on a few input-output demonstrations without any model training.
arXiv:2601. 03309v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models, which integrate pretrained large Vision-Language Models (VLM) into their policy backbone, are gaining significant attention for their promising generalization capabilities.
arXiv:2606. 29892v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become indispensable for pushing Vision-Language-Action Models (VLAs) beyond static imitation learning.
arXiv:2608.22301v1 Announce Type: cross Abstract: Humans imitate at the level of intent: given a demonstration, we infer its goal and carry it out with whatever tools, objects, and layouts are at han...
arXiv:2602. 19313v2 Announce Type: replace-cross Abstract: General-purpose robot learning requires dense, instruction-conditioned feedback that can distinguish meaningful task progress from stalled, failed, or partially completed behavior.
The paper introduces CRAFT, a method for improving compositional generalization in vision‑language‑action (VLA) models. It addresses the issue where models rely on visual shortcuts during fine‑tuning, leading them to execute demonstrated skill combinations that match observations rather than the instructed ones. By training with counterfactual instruction–observation pairs and transferring supervision through reusable skill representations, CRAFT enhances success on unseen skill combinations while preserving performance on demonstrated ones across multiple VLA models and benchmarks.
The paper introduces MIME, a task‑agnostic visual in‑context learning (vICL) dataset that removes textual shortcuts by omitting explicit rules linking demonstration images to answers, forcing models to infer the rule visually. Fine‑tuning vision‑language models on MIME boosts vICL accuracy by 8.1–18.0 points across three VLMs while preserving single‑image performance and improving multi‑image and text‑only in‑context learning. The authors also propose DTAD, a technique that randomly drops demonstration labels during training to encourage visual inference, achieving comparable accuracy with more consistent attention to demonstration images and outperforming frontier models on the TrueMICL benchmark.
The paper investigates integrating Test‑Time Training (TTT) into InternVLA‑A1 and Pi0.5 to enable robots to generate actions conditioned on human demonstrations. By training on paired demonstration and robot trajectory data, the authors explore memory transfer, truncated backpropagation, and contrastive regularization across video, short‑video, and single‑frame cue tasks. Experiments show that, at the current scale, adding TTT and paired training does not consistently improve target selection from demonstrations, and even single‑frame cues still lead robots to prefer cups at fixed positions.