A focused look at this week’s most significant advances in robotics & embodied ai \u2014 17 papers surveyed from arXiv and leading AI labs.<\/p>\n
—<\/p>\n
Embodied AI is having a moment. From dexterous manipulation to whole-body loco-manipulation, this week’s papers show robots learning more complex behaviors from less data.<\/p>\n
ForesightSafety-VLA: A Unified Diagnostic Safety Benchmark for Vision-Language-Action Models<\/strong> \u2014 Mingyang Lyu, Yinqian Sun, Yiyang Jia, Sicheng Shen, Moquan Sha, Huangrui Li, Feifei Zhao, Yi Zeng<\/em><\/p>\n In embodied intelligence, safety is a prerequisite for reliable robot deployment in the physical world. Current vision-language-action (VLA) models continue to advance toward general-purpose task capa…<\/p>\n arXiv<\/a><\/p>\n E-TTS: A New Embodied Test-Time Scaling Framework for Robotic Manipulation<\/strong> \u2014 Wen Ye, Peiyan Li, Tingyu Yuan, Yuan Xu, Xiangnan Wu, Chaoyang Zhao, Jing Liu, Nianfeng Liu, Yan Huang, Liang Wang<\/em><\/p>\n Recently, a few works have made early attempts to study test-time scaling for embodied tasks. However, two major challenges remain unsolved: (1) reasoning can effectively improve the performance of th…<\/p>\n arXiv<\/a><\/p>\n Scalable Behavior Cloning with Open Data, Training, and Evaluation<\/strong> \u2014 Arthur Allshire, Himanshu Gaurav Singh, Ritvik Singh, Adam Rashid, Hongsuk Choi, David McAllister, Justin Yu, Yiyuan Che<\/em><\/p>\n We introduce ABC, a fully open-source stack for manipulation with behavior cloning. At its core is ABC-130K: the largest open-source teleoperation dataset to date, featuring 3,500 hours of data spanni…<\/p>\n arXiv<\/a><\/p>\n LA4VLA: Learning to Act without Seeing via Language-Action Pretraining<\/strong><\/p>\n Tao Lin, Yuxin Du, Yiran Mao, Zewei Ye, Yilei Zhong, Bing Cheng, Yiming Wang, Jiting Liu, Yang Tian,<\/em><\/p>\n Vision-Language-Action (VLA) models are commonly pretrained on robot demonstrations by jointly mapping visual observations and language instructions to actions. However, dense visual-action supervision can dominate the comparatively sparse…<\/p>\n arXiv<\/a><\/p>\n RecallRisk-BERT: A Multi-Task Framework for Post-Report Medical Device Recall Triage<\/strong><\/p>\n Ali Semih Atalay, Sevgi Yigit-Sert<\/em><\/p>\n Medical device recalls are a critical regulatory mechanism for protecting patient safety. The growing volume of FDA recall records presents challenges in post-report recall triage, severity assessment, and root-cause interpretation. Existing…<\/p>\n arXiv<\/a><\/p>\n Advancing Omnimodal Embodied Agents from Isolated Skills to Everyday Physical Autonomy<\/strong><\/p>\n Junhao Shi, Zezheng Huai, Siyin Wang, Jia Chen, Yubang Wang, Zhaoye Fei, Hechang Chen, Jingjing Gong<\/em><\/p>\n Building persistent embodied agents in unstructured environments demands unified orchestration of heterogeneous tools spanning both cyber (APIs, IoT) and physical (manipulation, navigation) domains, coupled with autonomous recovery from physical…<\/p>\n arXiv<\/a><\/p>\n Learning to Fold: prizewinning solution at LeHome Challenge 2026 (1st place online, 2nd offline)<\/strong><\/p>\n Ilia Larchenko<\/em><\/p>\n I describe my solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding. The system placed 1st of 62 teams in the online (simulation) round and 2nd in the real-world final. It improves a vision-language-action…<\/p>\n arXiv<\/a><\/p>\n World Action Models Enable Continual Imitation Learning with Recurrent Generative Replays<\/strong><\/p>\n Manish Kumar Govind, Dominick Reilly, Smit Patel, Hieu Le, Srijan Das<\/em><\/p>\n Going beyond predicting robot actions, World Action Models (WAMs) can also generate future visual observations. We build on this generative capability to propose Recurrent Generative Replay (REGEN), a continual imitation learning framework that…<\/p>\n arXiv<\/a><\/p>\n OctoSense: Self-Supervised Learning for Multimodal Robot Perception<\/strong><\/p>\n Anthony Bisulco, Jeremy Wang, Kostas Daniilidis, Randall Balestriero, Pratik Chaudhari<\/em><\/p>\n We present OctoSense, an open-source sensor platform with stereo RGB and event cameras, LiDAR, a thermal camera, an inertial measurement unit, RTK-corrected global positioning system, and proprioception (CAN bus data from a car, and joint angles…<\/p>\n arXiv<\/a><\/p>\n RouterVLA: Turning Smoke Tests into Supervision for Heterogeneous VLA Selection<\/strong><\/p>\n Xingyu Ren, Chugang Yi, Ge Ma, Youran Sun<\/em><\/p>\n We study whether pre-deployment evaluation rollouts can be reused to supervise policy selection. Robot teams routinely smoke test candidate vision-language-action (VLA) policies, then compress those trials into a global winner. RouterVLA evaluates…<\/p>\n arXiv<\/a><\/p>\n VibeAct: Vibration to Actions for Contact-Rich Reactive Robot Dexterity<\/strong><\/p>\n Yuemin Mao, Uksang Yoo, Jean Oh, Jonathan Francis, Jeffrey Ichnowski<\/em><\/p>\n Dexterous manipulation depends on contact events that are fast, local, and often visually occluded. Piezoelectric microphones offer a compact and high-bandwidth way to sense these interactions, but the resulting vibro-acoustic signals are difficult…<\/p>\n arXiv<\/a><\/p>\n HumanoidUMI: Bridging Robot-Free Demonstrations and Humanoid Whole-Body Manipulation<\/strong><\/p>\n Hongwu Wang, Chenhao Yu, Youhao Hu, Jiachen Zhang, Yuanyuan Li, Shaqi Luo<\/em><\/p>\n High-quality demonstration data are essential for humanoid robot skill learning, especially for whole-body behaviors that require coordinated perception, locomotion, and manipulation. Existing data-collection methods largely rely on robot…<\/p>\n arXiv<\/a><\/p>\n PhysReflect-VLA: Physical Feasibility and Self-Reflective Regulation for Reliable Vision-Language-Action Policies<\/strong><\/p>\n Jiayu Yang, Tao Yang, Weijun Li, Xiang Chang, Fei Chao, Changjing Shang, Qiang Shen<\/em><\/p>\n Long-horizon robotic manipulation is highly sensitive to physically infeasible transitions, contact-induced disturbances, and the lack of effective self-correction during execution. Although Vision-Language-Action (VLA) models provide strong task…<\/p>\n arXiv<\/a><\/p>\n PAMAE: Phase-Aware-MoE Action Experts Towards Reliable Flow-Matching Vision-Language-Action Policies<\/strong><\/p>\n Jiayu Yang, Tao Yang, Xiang Chang, Fei Chao, Changjing Shang, Qiang Shen<\/em><\/p>\n Reliable action generation for multi-stage robotic manipulation remains challenging for Vision-Language-Action (VLA) models. While existing flow-matching VLA policies offer strong multimodal grounding and generalization, they typically employ a…<\/p>\n arXiv<\/a><\/p>\n Ordinal Neural Collapse as a Representation Prior for Visual Navigation<\/strong><\/p>\n E-In Son, Jung-Taak Kim, Seung-Woo Seo<\/em><\/p>\n Learning robust navigation policies directly from visual observations remains a fundamental challenge in vision-based robotic navigation. In end-to-end imitation learning approaches, the visual encoder and action decoder are jointly optimized using…<\/p>\n arXiv<\/a><\/p>\n Improving Vision-Language-Action Model Fine-Tuning with Structured Stage and Keyframe Supervision<\/strong><\/p>\n Yuan Xu, Yixiang Chen, Kai Wang, Jiabing Yang, Peiyan Li, Qisen Ma, Yan Huang, Liang Wang<\/em><\/p>\n Vision-Language-Action (VLA) models have shown strong potential for generalizable robotic manipulation. During fine-tuning, however, action supervision applies equally across all timesteps, without structured supervision on which manipulation stage…<\/p>\nTraining & Scaling<\/h2>\n
Safety & Alignment<\/h2>\n
Additional Research<\/h2>\n