# LIBERO

A frozen π0.5 policy gains a local CALL head that pages GPT-6 Astra only at critical recovery moments, raising success from 95.5% to 97.5% across 200 LIBERO episodes while confining Astra to a small share of environment steps.

A frozen π0.5 policy gains a local CALL head that pages GPT-6 Astra only at critical recovery moments, raising success from 95.5% to 97.5% across 200 LIBERO episodes while confining Astra to a small share of environment steps.

A pretrained vision-language model can act without robot-action pretraining once a tool-center-point action frame is projected into the wrist-camera view and actions are emitted through its native text interface, reaching 79.4% average zero-shot success on simple manipulation tasks, close to VLA models fine-tuned with about 30 demonstrations.

A pretrained vision-language model can act without robot-action pretraining once a tool-center-point action frame is projected into the wrist-camera view and actions are emitted through its native text interface, reaching 79.4% average zero-shot success on simple manipulation tasks, close to VLA models fine-tuned with about 30 demonstrations.

SmolWAM compresses the 2B-parameter Cosmos Policy video world-action model to 0.88B, succeeding in 95.5% of 400 LIBERO episodes at 57 ms, 2.3 TFLOP and 2.4 GiB per 16-action chunk, 11 times faster than its teacher while still forecasting scene motion.

SmolWAM compresses the 2B-parameter Cosmos Policy video world-action model to 0.88B, succeeding in 95.5% of 400 LIBERO episodes at 57 ms, 2.3 TFLOP and 2.4 GiB per 16-action chunk, 11 times faster than its teacher while still forecasting scene motion.

CAPABLE is a capability-aware adaptation framework for frozen vision-language-action policies: a temporal encoder shared across joints infers from command-response history and live Jacobian grounding how much commanded motion each joint still realizes, a self-supervised physical-prediction objective ties that representation to behavior, and a bounded residual reinforcement-learning policy corrects the VLA arm action without fault labels or faulty-joint identifiers; across 28 LIBERO tasks it raises success on an actuator excluded from fault training from 24.8% to 59.3%, 17.4 points above a parameter-matched global-history baseline, while preserving healthy performance.

CAPABLE is a capability-aware adaptation framework for frozen vision-language-action policies: a temporal encoder shared across joints infers from command-response history and live Jacobian grounding how much commanded motion each joint still realizes, a self-supervised physical-prediction objective ties that representation to behavior, and a bounded residual reinforcement-learning policy corrects the VLA arm action without fault labels or faulty-joint identifiers; across 28 LIBERO tasks it raises success on an actuator excluded from fault training from 24.8% to 59.3%, 17.4 points above a parameter-matched global-history baseline, while preserving healthy performance.

This work characterizes the language sensitivity of vision-language-action models (VLAs): one-word edits can move success by tens of points, and phrasing alone nearly closes the 21-point gap between in-distribution and out-of-distribution tasks; the authors have a large language model distill scored phrasings of a few training tasks into ten to twenty rephrasing rules, rewrite each incoming instruction once at deployment, and thereby improve the frozen policy by 16–27% relative on twelve held-out tasks without modifying its weights, lifting in-finetune success on LIBERO from 93.6% to 97.8%.

This work characterizes the language sensitivity of vision-language-action models (VLAs): one-word edits can move success by tens of points, and phrasing alone nearly closes the 21-point gap between in-distribution and out-of-distribution tasks; the authors have a large language model distill scored phrasings of a few training tasks into ten to twenty rephrasing rules, rewrite each incoming instruction once at deployment, and thereby improve the frozen policy by 16–27% relative on twelve held-out tasks without modifying its weights, lifting in-finetune success on LIBERO from 93.6% to 97.8%.

The work proposes a Bayesian non-parametric skill prior built on a Dirichlet Process Mixture that models temporally extended skills in a structured latent space, enabling adaptive skill discovery without predefining the number of components; integrated into a hierarchical RL framework, the learned prior guides high-level skill selection while a pretrained decoder generates temporally abstracted actions, and experiments on Franka Kitchen, LIBERO-Long, Meta-World, and a real robot show consistent gains in long-horizon manipulation, achieving over 0.8 success rate within 1.5M steps whereas SAC fails to converge even after 5M steps (below 0.1), with an average improvement of 21.8% over a single-Gaussian prior baseline.

The work proposes a Bayesian non-parametric skill prior built on a Dirichlet Process Mixture that models temporally extended skills in a structured latent space, enabling adaptive skill discovery without predefining the number of components; integrated into a hierarchical RL framework, the learned prior guides high-level skill selection while a pretrained decoder generates temporally abstracted actions, and experiments on Franka Kitchen, LIBERO-Long, Meta-World, and a real robot show consistent gains in long-horizon manipulation, achieving over 0.8 success rate within 1.5M steps whereas SAC fails to converge even after 5M steps (below 0.1), with an average improvement of 21.8% over a single-Gaussian prior baseline.