← All threads

Robot Manipulation Policies

Methods for training and augmenting vision-language-action policies for robot manipulation, including affordance grounding and architectural design.

31 papers

Where this stands

The written synthesis of this thread is for subscribers. Subscribe.

Results across this thread

3 reported results from the papers in this thread.

PaperBenchmarkMetricResultConditionsCode

The table is for subscribers. Subscribe to see every reported number side by side.

How this thread developed

  1. Oct 2026 · HKU, CUHK(SZ)

    Robot policies encode changed instructions but still follow scene cues

    Tests whether language-conditioned robot policies actually follow instructions vs scene cues, revealing grounding issues.

  2. Oct 2026 · Stanford University

    Guided by failures, choosing initial states improves robot learning per supervised episode

    Introduces a failure-guided initial state selection strategy that improves robot learning data efficiency.

  3. 4 further papers

    Oct 2026 · POSTECH, KAIST

    Forecasting 3D scene motion improves dexterous robot manipulation

    introduces a policy pretrained on human demonstrations using 3D point trajectory forecasting, achieving better performance on dexterous tasks.

  4. 4 further papers

    Oct 2026 · Tsinghua University, Chongqing University

    Retrieval replaces generation for building simulation-ready articulated objects from a single photo

    Constructs physically valid articulated simulation assets from a single image and task instruction, enabling affordance grounding for robot manipulation policies without generation.

  5. 3 further papers

    Oct 2026 · University of Washington, NVIDIA

    Adaptive sampling helps million-environment robot training solve harder tasks

    Adaptive sampling in massively parallel robot RL improves exploration efficiency for harder tasks.

  6. Oct 2026 · Massachusetts Institute of Technology, Amazon

    Safety filter separates task and safety learning in robot control

    Introduces a safety filter that decouples task and safety learning in robot control, advancing safe policy training.

  7. 2 further papers

    Oct 2026 · NVIDIA, MIT

    Action-aware embeddings retrieve better robot experiences for future actions

    Introduces a retrieval-based method for robot experience representation using vision-language-action embeddings.

  8. 2 further papers

    Oct 2026 · Kyutai, Inria

    Compact robot policy uses lightweight memory to stabilize action generation

    Shows that a lightweight memory buffer stabilizes action generation in a compact vision-language-action model.

  9. Oct 2026 · Amazon, University of California, San Diego

    Humanoid repositions bulky objects using its arms and torso

    Treats humanoid arms and torso as coordinated contacts for whole-body manipulation of bulky objects.

  10. 4 further papers

    Oct 2026 · Zhejiang University, Ant Group

    BiDex-100 tests robot policies across 106 coordinated two-handed tasks

    Adds BiDex-100, a 106 task benchmark for two multi-fingered hands with out-of-domain variants to evaluate bimanual manipulation policies.

  11. Oct 2026 · Brown University, Amazon

    Generated language gives robot policies a working memory

    Integrates video, action, and language in one architecture to give robot policies working memory through generated language.

  12. Oct 2026 · Carnegie Mellon University, NVIDIA

    One Human Demonstration Can Seed Adaptable Dexterous Robot Skills

    Learns dexterous manipulation from a single human demonstration by combining imitation and reaching goals.

12 of 31 papers shown