Research Papers Topics
Models that map perception and language directly to robot actions.
Vision-language-action models try to make robots learn from the same semantic world that language and vision models already use. The goal is not only recognizing objects or parsing instructions, but producing continuous or tokenized actions that work on real hardware.
This topic is early but strategically important. RT-2 shows how web-scale vision-language knowledge can transfer into robot control by representing actions in a language-like format. π0 pushes toward a more general robot policy with flow matching over continuous actions. The hard open questions are data collection, safety, embodiment transfer, latency, and whether a single policy can handle the physical diversity of real robots.
Vision-Language-Action · Physical Intelligence
π0 bolts a flow-matching action expert onto a pretrained VLM, emitting ~50Hz action chunks so one policy can fold laundry, bus tables, and assemble boxes across single-arm, dual-arm, and mobile robots.
Vision-Language-Action · Google DeepMind
RT-2 co-fine-tunes a web-pretrained vision-language model on robot trajectories, expresses actions as text tokens, and gets emergent generalization to novel objects, unseen commands, and basic reasoning across 6k trials.
Vision-Language-Action · Google DeepMind
RT-2 co-fine-tunes a web-pretrained vision-language model on robot trajectories, expresses actions as text tokens, and gets emergent generalization to novel objects, unseen commands, and basic reasoning across 6k trials.
Vision-Language-Action · Physical Intelligence
π0 bolts a flow-matching action expert onto a pretrained VLM, emitting ~50Hz action chunks so one policy can fold laundry, bus tables, and assemble boxes across single-arm, dual-arm, and mobile robots.
Vision-Language-Action · Allen Institute for AI
MolmoAct2 is an open vision-language-action stack that reasons in 3D before acting. On real-world DROID it hits 87.1% success, +38.7 points over the runner-up, and its Molmo2-ER brain beats GPT-5 and Gemini Robotics ER.
Vision-Language-Action · RLWRLD
RLDX-1, from RLWRLD and KAIST, adds motion, memory and tactile streams to a Qwen3-VL backbone. It catches fast-moving objects 87.5% of the time vs 29.2% for pi0.5, and beats GR00T N1.6 on LIBERO-Plus 86.7% to 72.6%.
DragMesh-2 opens doors and drawers with a 51-DoF hand and no actuator on the object joint, so motion comes only from contact. PICA training hits 0.89 success at nominal damping and 0.56 at 4x, with no tactile sensing.
Vision-Language-Action · CASIA
World Pilot adds two world-model priors to a VLA policy and hits 84.7% total success on LIBERO-Plus zero-shot OOD, up 4.2 over the ABot-M0 base, and the scene prior works even from a world model never action-trained.
Vision-Language-Action · ACE Robotics
ACE-Ego-0 pretrains a VLA on 6,000+ hours mixing robot trajectories with human egocentric video turned into pseudo-actions. It averages 78.3% on six real bimanual tasks vs 71.7% for pi-0.5 and 35.6% for GR00T-N1.7.
Vision-Language-Action · X Square Robot
WALL-WM organizes VLA pretraining around semantic action events, not fixed-length chunks. Its event mode scores 75.86 Task Progress on diverse real-robot manipulation versus 55.64 for pi0.5.
Vision-Language-Action · Zhejiang University
LabVLA trains a Qwen3-VL-4B backbone plus DiT action expert on laboratory workflows and reports 71.1% ID and 70.0% OOD success on LabUtopia.
World Models · Independent Researcher
AnchorWorld: Egocentric World Simulation for Embodied AI turns egocentric world simulation into a checkable test, with concrete failure signals, benchmark limits, and builder takeaways.
DragMesh-2 opens doors and drawers with a 51-DoF hand and no actuator on the object joint, so motion comes only from contact. PICA training hits 0.89 success at nominal damping and 0.56 at 4x, with no tactile sensing.
Vision-Language-Action · CASIA
World Pilot adds two world-model priors to a VLA policy and hits 84.7% total success on LIBERO-Plus zero-shot OOD, up 4.2 over the ABot-M0 base, and the scene prior works even from a world model never action-trained.
Vision-Language-Action · ACE Robotics
ACE-Ego-0 pretrains a VLA on 6,000+ hours mixing robot trajectories with human egocentric video turned into pseudo-actions. It averages 78.3% on six real bimanual tasks vs 71.7% for pi-0.5 and 35.6% for GR00T-N1.7.
Vision-Language-Action · X Square Robot
WALL-WM organizes VLA pretraining around semantic action events, not fixed-length chunks. Its event mode scores 75.86 Task Progress on diverse real-robot manipulation versus 55.64 for pi0.5.
Vision-Language-Action · Zhejiang University
LabVLA trains a Qwen3-VL-4B backbone plus DiT action expert on laboratory workflows and reports 71.1% ID and 70.0% OOD success on LabUtopia.
World Models · Independent Researcher
AnchorWorld: Egocentric World Simulation for Embodied AI turns egocentric world simulation into a checkable test, with concrete failure signals, benchmark limits, and builder takeaways.
Robotics · Independent Researcher
TVRBench: Can Models Move to a Target Viewpoint? turns active 3D viewpoint reproduction into a checkable test, with concrete failure signals, benchmark limits, and builder takeaways.
Robotics · Tsinghua University
Humanoid-GPT treats humanoid control like language modeling: a causal Transformer distilled from ~384 PPO experts on a 2-billion-frame corpus, 200x prior data. It hits 92.58 percent sim success, under 1.5ms.
AI Agents · Shanghai Jiao Tong University
MMSkills packages textual procedures, runtime state cards, and keyframes into reusable skills for visual agents, lifting Qwen3-VL-235B from 21.34% to 39.17% on OSWorld and a small 8B model from 10.78% to 25.40%.
Vision-Language-Action · Allen Institute for AI
MolmoAct2 is an open vision-language-action stack that reasons in 3D before acting. On real-world DROID it hits 87.1% success, +38.7 points over the runner-up, and its Molmo2-ER brain beats GPT-5 and Gemini Robotics ER.
Vision-Language-Action · Shanghai AI Laboratory
PhysBrain 1.0 compiles human egocentric video into physics QA to pretrain a VLM, then adapts it to robot control — lifting Franka grasping from 47.1% to 63.3% over 50 trials versus a pi0.5 baseline.
Vision-Language-Action · Alibaba Qwen Team
Qwen-VLA extends Qwen's vision-language stack with a DiT action decoder and embodiment-aware prompts to run manipulation, navigation, and trajectory prediction in one model — 97.9% on LIBERO and 69.0% OSR on R2R.
Vision-Language-Action · RLWRLD
RLDX-1, from RLWRLD and KAIST, adds motion, memory and tactile streams to a Qwen3-VL backbone. It catches fast-moving objects 87.5% of the time vs 29.2% for pi0.5, and beats GR00T N1.6 on LIBERO-Plus 86.7% to 72.6%.
Vision-Language-Action · ETH Zurich
A position paper from ETH Zurich, Stanford and TU Darmstadt argues scaling VLA and world models is not enough — robots need four interfaces to turn unstructured human and video behaviour into grounded supervision.
Vision-Language-Action · Physical Intelligence
π0 bolts a flow-matching action expert onto a pretrained VLM, emitting ~50Hz action chunks so one policy can fold laundry, bus tables, and assemble boxes across single-arm, dual-arm, and mobile robots.
Vision-Language-Action · Google DeepMind
RT-2 co-fine-tunes a web-pretrained vision-language model on robot trajectories, expresses actions as text tokens, and gets emergent generalization to novel objects, unseen commands, and basic reasoning across 6k trials.