Technology areas
Robot models and learning
Robot VLA models, general action and world models, imitation and reinforcement learning, skill and motion policy learning, online adaptation, and transfer across tasks and robot bodies. The contribution is a model, policy, or learning method. Dataset releases belong under Data and simulation; inference deployment belongs under Software and compute systems.
206 selected · 116 in the last 30 days
arXiv cs.RO ·
Paper analyzes agentic manipulation policies: AMP1 achieves 85.6% zero-shot success rate
The paper proposes agentic manipulation policies (AMP), in which a generalist AI model directly controls a robotic arm and adjusts its actions as the task progresses. The authors' reference implementation, AMP1, runs unchanged across single-arm, bimanual, and mobile manipulation.
arXiv cs.RO ·
A2-MEAS: An Embodied Multi-Agent Framework with Code as a Scaling Interface
The authors propose A2-MEAS, an embodied multi-agent framework that uses code as a scaling interface, combining corrective language, spatial evidence, execution history, and action proposals into correction codes; during training, MetaBrain provides supervision and runtime guidance.
机器人前瞻 · 微信公众号 ·
NVIDIA releases VFlash visual planning decoding framework, robot action decoding up to 12x faster
NVIDIA has released VFlash, a general-purpose decoding framework for fast visual planning that replaces token-by-token coordinate generation with parallel decoding of complete visual primitives, allowing it to predict robot grasp positions and motion trajectories in one pass. At batch size 1 on H100, single-generation speed improved by about 25x; across four MolmoSpaces tasks, the world action model averaged a 70.2% success rate versus a 62.6% baseline. Jim Fan shared that it achieves up to 12x faster action decoding, 5.1x overall robot inference speedup, and up to 61% lower 3D trajectory error.
至简动力 · 微信公众号 ·
Zhijian Dynamics and Hangcha Group deploy robot sorting line for supermarket front warehouses
On November 3, Zhijian Dynamics and Hangcha Group completed the deployment of the first robot sorting line for a supermarket front warehouse, connecting the full workflow from AMR merge feeding, robot autonomous planning and dexterous grasping and turning, and scan verification to order distribution. The line uses the iX robot body that Zhijian Dynamics customized for Hangcha, and relies on visual perception and VLA task planning to handle boxed, bagged, bottled, and other product forms without the need to model each SKU separately. Measured single-machine sorting efficiency reaches up to 450 items per hour, with accuracy stable above 98%, and under full load it can support 24-hour continuous fault-free operation.
Xbotics具身智能实验室 · 微信公众号 ·
GE-Sim 2.0 and π0.5 closed-loop simulation test: Why does reward hacking occur in VLA training?
The Xbotics Embodied Intelligence Lab built a closed-loop simulation pipeline using the AgiBot GE-Sim 2.0 world model and the OpenPI π0.5 policy. In testing, the world model generated about 4.33 frames per second, the pure-policy closed loop ran at about 3.20 frames per second, and after adding MiMo online scoring, the rate dropped to about 0.88 frames per second.
PsiBot/灵初智能 ·
Lingchu Intelligence and collaborators propose FAFM, a continuous action generation method for heterogeneous-frequency data, accepted to NeurIPS 2026
Lingchu Intelligence, together with Beihang University, Peking University, and The Chinese University of Hong Kong, proposed a frequency-aware flow matching method called FAFM; the paper was accepted to the NeurIPS 2026 main conference. The method uses the discrete cosine transform (DCT-II) to represent actions as frequency-domain coefficients corresponding to real time, performs flow matching in coefficient space and constrains the first-order time derivative, requires no additional learnable parameters, and can be used independently or integrated into VLA models such as π₀ and π₀.₅.
PsiBot/灵初智能 ·
PKU-Spirit AI Joint Lab's RoboFFT Accepted to CoRL2026, Uses Reinforcement Learning Post-Training to Generate Robotic Policies
RoboFFT, proposed by the PKU-Spirit AI Joint Lab, has been accepted to CoRL2026. It is a reinforcement learning post-training method for generative robotic policies that reuses the forward noising process from diffusion model and flow model pretraining to construct an approximate probability ratio for PPO-style clipped updates.
3D视觉工坊 · 微信公众号 ·
Any3R goes open source: one model unifies 3D reconstruction and pose estimation for pinhole, fisheye, and panoramic cameras
South Dakota State University and other institutions propose Any3R, which uses a dense ray field to describe arbitrary central cameras, reconstructing a 3D point cloud and estimating pose from mixed-camera images in a single forward pass. The method replaces the row and column indices of rotary position encoding with angular coordinates and weights the pixel loss by solid angle, so one set of weights covers pinhole, fisheye beyond 180°, and 360° panoramic cameras. On benchmarks spanning 13 central camera types and 3 datasets, it leads in both reconstruction accuracy and pose accuracy; the inference code, 13 camera models, and weights are open-sourced.
arXiv cs.RO ·
TacClaw-Infinity: A Large-Scale Vision-Tactile-Language-Action Dataset for Dexterous Manipulation
The research team released TacClaw-Infinity, a large-scale structured Vision-Tactile-Language-Action (VTLA) dataset for contact-rich manipulation, spanning over 10,000 hours, 1 million episodes, and 1 billion frames across more than 1,000 task types.
arXiv cs.RO ·
Movens-Zero: Scaling humanoid motion foundation models with large-scale human videos
The research team proposes Movens-Zero, a scalable understanding-generation-execution architecture that supports real-time closed-loop humanoid motion generation under text, speech, music, and video inputs. Its accompanying dataset, Movens-5M, is built from Internet videos through clip generation, quality filtering, semantic annotation, motion recovery, and physical refinement with humanoid retargeting. It contains 15 million human motion sequences and 5 million physically executable humanoid motion sequences, totaling over 10,000 hours.
arXiv cs.RO ·
SimuGuard: Auditing Contact Artifacts in Simulation Benchmarks Such as LIBERO and RoboTwin
The authors propose SimuGuard, an auditing plugin for LIBERO, RoboTwin, RoboCasa, and RoboDojo that identifies failures caused by the simulator rather than the policy. On RoboTwin, the authors trace these failures to the physics engine's collision detection: when geometry is 1 to 2 centimeters apart, the engine reports a penetration of about 2 centimeters, and the solver ejects the object at several meters per second because the distance algorithm discards triangles below a fixed area threshold.
arXiv cs.RO ·
∇Mimic: Efficient Whole-Body Control for Humanoid Robots with Differentiable Simulation
The authors propose ∇Mimic, which uses exact gradients from a differentiable simulator to train whole-body motion-tracking policies for humanoid robots, reaching a score of 0.82 in 20 minutes and 0.874 in 1.8 hours on a single GPU, whereas the world-model method takes 1.2 hours to reach 0.82.
arXiv cs.RO ·
From Teleoperation to Long-Horizon Dexterity: Evaluating Visuomotor Skills and Offline Proxies for Dexterous Manipulation
The research team used a five-fingered robotic hand to grasp and insert RJ45 connectors, evaluating 281 imitation-learning policies spanning seven policy families in approximately 1,700 real-robot trials. Controlled ablations showed that observation and action representations had a greater impact than conventional optimizer and sampling choices. For point-cloud policies, region-of-interest cropping, representation in the robot-base frame, and normalization within the point encoder each improved success by 70 to 100 percentage points. For the vision-language-action policy, switching to absolute action prediction improved insertion success by up to 70 percentage points.
甲子光年 · 微信公众号 ·
Peking University's Dong Hao team proposes One Scan is Enough: embodied navigation post-training with a single scan
Dong Hao's team at Peking University has published the paper "One Scan is Enough," which uses a LiDAR-equipped iPhone to scan a target environment once, automatically generates navigation tasks, and uses the reconstructed scene geometry for reinforcement post-training, without needing manual teleoperation demonstrations.
机器之心 · 微信公众号 ·
Fysics AI and Fudan propose Fysiverse-UniWAM unified world-action model, ranks first in both navigation and mobile manipulation across 24 real-robot tests
Fysics AI, together with Fudan University's CITLab, proposed the unified world-action model Fysiverse-UniWAM, which uses a single model to handle autonomous navigation, mobile manipulation, and fixed-workstation dexterous manipulation.
3D视觉工坊 · 微信公众号 ·
HKU open-sources GenSfM: jointly reconstructing 3D objects and camera poses from four uncalibrated images
A University of Hong Kong team has open-sourced GenSfM, which uses a native 3D diffusion model to jointly denoise object shape and per-view camera poses within a single multimodal sequence, requiring only 2 to 4 uncalibrated images and no camera intrinsics or extrinsics. Evaluated on Toys4k and GSO, it improves novel-view PSNR by 3.2–4.5 dB over the multi-view baseline TRELLIS-M, reduces median Chamfer distance by about 64%–74%, and achieves pose Acc@5° above 96%.
机器人前瞻 · 微信公众号 ·
AgiBot releases WITA-OMNI interaction model and Lingqu OS 1.0, full open source by year-end
At its 2026 Innovation Day, AgiBot released WITA-OMNI, an embodied omni-modal end-to-end interaction large model. It ranked first on four benchmarks: Daily-Omni, TimeLens-Bench, EmbSpatialBench, and MMAU, with an average response time of about 600 milliseconds.
机器之心 · 微信公众号 ·
Qiyuan releases PrimeSim world-model simulator, tops EWMBench and WorldSimProbe
The Qiyuan Embodied Research Center has released PrimeSim, a world-model simulator for robot manipulation tasks, built by adapting a pretrained video generation model through mid-training and post-training stages. It ranks first on EWMBench with an overall score of 6.09, and took first place in three of five WorldSimProbe evaluations: global trajectory coverage, action-source behavior preservation, and interaction dynamics.
机器之心 · 微信公众号 ·
PokeBot releases embodied API RoboAPI, world-action model Poke-Aurora debuts
PokeBot released its first model product, RoboAPI, in the early hours of October 30. It is a publicly accessible, out-of-the-box embodied API: after developers connect and adapt a robotic arm, they need neither to collect data nor train models, and can issue commands with a single prompt to execute tasks the model was not specifically trained on.
3D视觉工坊 · 微信公众号 ·
Infinigence AI, Tsinghua and AgiBot propose Collective Experience Learning (CEL), validated on 96 robots with 98.3% success rate
Infinigence AI, together with Tsinghua University and AgiBot, proposed the Collective Experience Learning (CEL) paradigm and completed prototype validation on 96 robots. In a phone packaging task, policies trained with CEL completed 120 tests on two robots of the same model that did not participate in training, with the success rate rising from 79.2% to 98.3%; when the number of robots participating in learning increased from 16 to 96, the test success rate rose from 86.7% to 98.3%.