近三年论文 · 11 篇 （点击展开摘要，时间倒序）
RECALL: Recovery Experience Collection for Active Lifelong Learning in Vision-Language-Action Models
Vision-Language-Action (VLA) models are commonly fine-tuned through passive imitation learning, where additional demonstrations are collected for tasks where the policy performs poorly. This approach incurs several downsides: it requires the robot to fail before data collection is triggered, provides little guidance about which states require supervision, and wastes demonstrator effort on redundant parts of the task where the policy already performs well. In this paper, we propose an active, continual learning paradigm for VLAs. We demonstrate that active, uncertainty-guided data collection leads to more efficient fine-tuning than when using passively-collected demonstrations. However, we also find that fine-tuning only on actively-collected recovery data leads to catastrophic forgetting. We evaluate techniques for continual learning, including replay-based data mixing and elastic weight consolidation, and identify tradeoffs between plasticity to uncertainty-guided recovery data and retention of previously learned behaviors. Overall, our work contributes an empirical study of active continual learning for autoregressive VLAs, establishing that uncertainty-guided recovery demonstrations can improve adaptation efficiency while also revealing open challenges when targeted new data is incorporated into large robot policies.
RECALL: Recovery Experience Collection for Active Lifelong Learning in Vision-Language-Action Models
arXiv (Cornell University) · 2026 · cited 0
Vision-Language-Action (VLA) models are commonly fine-tuned through passive imitation learning, where additional demonstrations are collected for tasks where the policy performs poorly. This approach incurs several downsides: it requires the robot to fail before data collection is triggered, provides little guidance about which states require supervision, and wastes demonstrator effort on redundant parts of the task where the policy already performs well. In this paper, we propose an active, continual learning paradigm for VLAs. We demonstrate that active, uncertainty-guided data collection leads to more efficient fine-tuning than when using passively-collected demonstrations. However, we also find that fine-tuning only on actively-collected recovery data leads to catastrophic forgetting. We evaluate techniques for continual learning, including replay-based data mixing and elastic weight consolidation, and identify tradeoffs between plasticity to uncertainty-guided recovery data and retention of previously learned behaviors. Overall, our work contributes an empirical study of active continual learning for autoregressive VLAs, establishing that uncertainty-guided recovery demonstrations can improve adaptation efficiency while also revealing open challenges when targeted new data is incorporated into large robot policies.
The Simons Collaboration on Ecological Neuroscience: Studying how the brain interacts with the world.
The Simons Collaboration on Ecological Neuroscience: Studying how the brain interacts with the world
The Simons Collaboration on Ecological Neuroscience (SCENE) seeks to uncover general principles of brain function through an ecological perspective: studying perception, cognition, and action in the context of the affordances available to different agents. Here, we introduce SCENE's goals, hypotheses, and approaches outlining a collaborative vision for the next decade.
Enhancing Goal Inference via Correction Timing
Corrections offer a natural modality for people to provide feedback to a robot, by (i) intervening in the robot's behavior when they believe the robot is failing (or will fail) the task objectives and (ii) modifying the robot's behavior to successfully fulfill the task. Each correction offers information on what the robot should and should not do, where the corrected behavior is more aligned with task objectives than the original behavior. Most prior work on learning from corrections involves interpreting a correction as a new demonstration (consisting of the modified robot behavior), or a preference (for the modified trajectory compared to the robot's original behavior). However, this overlooks one essential element of the correction feedback, which is the human's decision to intervene in the robot's behavior in the first place. This decision can be influenced by multiple factors including the robot's task progress, alignment with human expectations, dynamics, motion legibility, and optimality. In this work, we investigate whether the timing of this decision can offer a useful signal for inferring these task-relevant influences. In particular, we investigate three potential applications for this learning signal: (1) identifying features of a robot's motion that may prompt people to correct it, (2) quickly inferring the final goal of a human's correction based on the timing and initial direction of their correction motion, and (3) learning more precise constraints for task objectives. Our results indicate that correction timing results in improved learning for the first two of these applications. Overall, our work provides new insights on the value of correction timing as a signal for robot learning.
Enhancing Goal Inference via Correction Timing
Corrections offer a natural modality for people to provide feedback to a robot, by (i) intervening in the robot's behavior when they believe the robot is failing (or will fail) the task objectives and (ii) modifying the robot's behavior to successfully fulfill the task. Each correction offers information on what the robot should and should not do, where the corrected behavior is more aligned with task objectives than the original behavior. Most prior work on learning from corrections involves interpreting a correction as a new demonstration (consisting of the modified robot behavior), or a preference (for the modified trajectory compared to the robot's original behavior). However, this overlooks one essential element of the correction feedback, which is the human's decision to intervene in the robot's behavior in the first place. This decision can be influenced by multiple factors including the robot's task progress, alignment with human expectations, dynamics, motion legibility, and optimality. In this work, we investigate whether the timing of this decision can offer a useful signal for inferring these task-relevant influences. In particular, we investigate three potential applications for this learning signal: (1) identifying features of a robot's motion that may prompt people to correct it, (2) quickly inferring the final goal of a human's correction based on the timing and initial direction of their correction motion, and (3) learning more precise constraints for task objectives. Our results indicate that correction timing results in improved learning for the first two of these applications. Overall, our work provides new insights on the value of correction timing as a signal for robot learning.
Enhancing Goal Inference via Correction Timing
arXiv (Cornell University) · 2026 · cited 0
Corrections offer a natural modality for people to provide feedback to a robot, by (i) intervening in the robot's behavior when they believe the robot is failing (or will fail) the task objectives and (ii) modifying the robot's behavior to successfully fulfill the task. Each correction offers information on what the robot should and should not do, where the corrected behavior is more aligned with task objectives than the original behavior. Most prior work on learning from corrections involves interpreting a correction as a new demonstration (consisting of the modified robot behavior), or a preference (for the modified trajectory compared to the robot's original behavior). However, this overlooks one essential element of the correction feedback, which is the human's decision to intervene in the robot's behavior in the first place. This decision can be influenced by multiple factors including the robot's task progress, alignment with human expectations, dynamics, motion legibility, and optimality. In this work, we investigate whether the timing of this decision can offer a useful signal for inferring these task-relevant influences. In particular, we investigate three potential applications for this learning signal: (1) identifying features of a robot's motion that may prompt people to correct it, (2) quickly inferring the final goal of a human's correction based on the timing and initial direction of their correction motion, and (3) learning more precise constraints for task objectives. Our results indicate that correction timing results in improved learning for the first two of these applications. Overall, our work provides new insights on the value of correction timing as a signal for robot learning.
INSIGHT: INference-time Sequence Introspection for Generating Help Triggers in Vision-Language-Action Models
Recent Vision-Language-Action (VLA) models show strong generalization capabilities, yet they lack introspective mechanisms for anticipating failures and requesting help from a human supervisor. We present \textbf{INSIGHT}, a learning framework for leveraging token-level uncertainty signals to predict when a VLA should request help. Using $π_0$-FAST as the underlying model, we extract per-token \emph{entropy}, \emph{log-probability}, and Dirichlet-based estimates of \emph{aleatoric and epistemic uncertainty}, and train compact transformer classifiers to map these sequences to help triggers. We explore supervision regimes for strong or weak supervision, and extensively compare them across in-distribution and out-of-distribution tasks. Our results show a trade-off: strong labels enable models to capture fine-grained uncertainty dynamics for reliable help detection, while weak labels, though noisier, still support competitive introspection when training and evaluation are aligned, offering a scalable path when dense annotation is impractical. Crucially, we find that modeling the temporal evolution of token-level uncertainty signals with transformers provides far greater predictive power than static sequence-level scores. This study provides the first systematic evaluation of uncertainty-based introspection in VLAs, opening future avenues for active learning and for real-time error mitigation through selective human intervention.
Effects of Robot Competency and Motion Legibility on Human Correction Feedback
As robot deployments become more commonplace, people are likely to take on the role of supervising robots (i.e., correcting their mistakes) rather than directly teaching them. Prior works on Learning from Corrections (LfC) have relied on three key assumptions to interpret human feedback: (1) people correct the robot only when there is significant task objective divergence; (2) people can accurately predict if a correction is necessary; and (3) people trade off precision and physical effort when giving corrections. In this work, we study how two key factors (robot competency and motion legibility) affect how people provide correction feedback and their implications on these existing assumptions. We conduct a user study <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(N=60)$</tex> under an LfC setting where participants supervise and correct a robot performing pick-and-place tasks. We find that people are more sensitive to suboptimal behavior by a highly competent robot compared to an incompetent robot when the motions are legible <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(p=0.0015)$</tex> and predictable <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(p=0.0055)$</tex>. In addition, people also tend to withhold necessary corrections <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(p < 0.0001)$</tex> when supervising an incompetent robot and are more prone to offering unnecessary ones <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(p=0.0171)$</tex> when supervising a highly competent robot. We also find that physical effort positively correlates with correction precision, providing empirical evidence to support this common assumption. We also find that this correlation is significantly weaker for an incompetent robot with legible motions than an incompetent robot with predictable motions <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(p=0.0075)$</tex>. Our findings offer insights for accounting for competency and legibility when designing robot interaction behaviors and learning task objectives from corrections.
Effects of Robot Competency and Motion Legibility on Human Correction Feedback
As robot deployments become more commonplace, people are likely to take on the role of supervising robots (i.e., correcting their mistakes) rather than directly teaching them. Prior works on Learning from Corrections (LfC) have relied on three key assumptions to interpret human feedback: (1) people correct the robot only when there is significant task objective divergence; (2) people can accurately predict if a correction is necessary; and (3) people trade off precision and physical effort when giving corrections. In this work, we study how two key factors (robot competency and motion legibility) affect how people provide correction feedback and their implications on these existing assumptions. We conduct a user study ($N=60$) under an LfC setting where participants supervise and correct a robot performing pick-and-place tasks. We find that people are more sensitive to suboptimal behavior by a highly competent robot compared to an incompetent robot when the motions are legible ($p=0.0015$) and predictable ($p=0.0055$). In addition, people also tend to withhold necessary corrections ($p < 0.0001$) when supervising an incompetent robot and are more prone to offering unnecessary ones ($p = 0.0171$) when supervising a highly competent robot. We also find that physical effort positively correlates with correction precision, providing empirical evidence to support this common assumption. We also find that this correlation is significantly weaker for an incompetent robot with legible motions than an incompetent robot with predictable motions ($p = 0.0075$). Our findings offer insights for accounting for competency and legibility when designing robot interaction behaviors and learning task objectives from corrections.
TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models
Existing benchmarks often highlight the remarkable performance achieved by state-of-the-art Multimodal Foundation Models (MFMs) in leveraging temporal context for video understanding. However, how well do the models truly perform visual temporal reasoning? Our study of existing benchmarks shows that this capability of MFMs is likely overestimated as many questions can be solved by using a single, few, or out-of-order frames. To systematically examine current visual temporal reasoning tasks, we propose three principles with corresponding metrics: (1) Multi-Frame Gain, (2) Frame Order Sensitivity, and (3) Frame Information Disparity. Following these principles, we introduce TOMATO, Temporal Reasoning Multimodal Evaluation, a novel benchmark crafted to rigorously assess MFMs' temporal reasoning capabilities in video understanding. TOMATO comprises 1,484 carefully curated, human-annotated questions spanning six tasks (i.e., action count, direction, rotation, shape & trend, velocity & frequency, and visual cues), applied to 1,417 videos, including 805 self-recorded and -generated videos, that encompass human-centric, real-world, and simulated scenarios. Our comprehensive evaluation reveals a human-model performance gap of 57.3% with the best-performing model. Moreover, our in-depth analysis uncovers more fundamental limitations beyond this gap in current MFMs. While they can accurately recognize events in isolated frames, they fail to interpret these frames as a continuous sequence. We believe TOMATO will serve as a crucial testbed for evaluating the next-generation MFMs and as a call to the community to develop AI systems capable of comprehending human world dynamics through the video modality.