Movens-Zero: Scaling humanoid motion foundation models with large-scale human video
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 humanoid retargeting physical refinement, and contains 15 million human motion sequences and 5 million physically executable humanoid motion sequences, with a total duration exceeding 10,000 hours.
Highlights
TacClaw-Infinity: A large-scale vision-tactile-language-action dataset for dexterous manipulation
The research team releases TacClaw-Infinity, a large-scale structured vision-tactile-language-action (VTLA) dataset for contact-intensive manipulation, with a scale exceeding 10,000 hours, 1 million episodes, and 1 billion frames, covering more than 1,000 task types.
arXiv cs.ROSource ↗Any3R open-sourced: 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 and can reconstruct a 3D point cloud and estimate poses 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, enabling a single set of weights to cover pinhole, fisheye beyond 180°, and 360° panoramic cameras. On benchmarks across 13 central camera types and 3 datasets, it leads in both reconstruction accuracy and pose accuracy, and its inference code, 13 camera models, and weights have been open-sourced.
3D视觉工坊 · 微信公众号Source ↗Walmart and Symbotic expand commercial agreement to add automation capabilities to 25 U.S. regional distribution centers
Walmart and Symbotic announced an expansion of their commercial agreement that will bring additional automation capabilities to 25 U.S. regional distribution centers and support further opportunities across Walmart's global supply chain network. The agreement includes an initial investment commitment of approximately $6 billion, covering multi-site, multi-year work. Symbotic says its integrated technologies, such as its BreakPack each-picking solution and SymMicro micro-fulfillment platform, are supported by AI software for scheduling and optimizing the flow of millions of cases per day.
Industry news
Walmart and Symbotic expand commercial agreement to add automation capabilities to 25 U.S. regional distribution centers
Walmart and Symbotic announced an expansion of their commercial agreement that will bring additional automation capabilities to 25 U.S. regional distribution centers and support further opportunities across Walmart's global supply chain network. The agreement includes an initial investment commitment of approximately $6 billion, covering multi-site, multi-year work. Symbotic says its integrated technologies, such as its BreakPack each-picking solution and SymMicro micro-fulfillment platform, are supported by AI software for scheduling and optimizing the flow of millions of cases per day.
Churchill Capital Corp XI shareholders approve merger with Agility Robotics
Churchill Capital Corp XI shareholders approved the business combination with Agility Robotics with more than 84% of votes in favor, and the transaction is expected to close on November 4.
Paper research
Movens-Zero: Scaling humanoid motion foundation models with large-scale human video
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 humanoid retargeting physical refinement, and contains 15 million human motion sequences and 5 million physically executable humanoid motion sequences, with a total duration exceeding 10,000 hours.
arXiv cs.ROSource ↗TacClaw-Infinity: A large-scale vision-tactile-language-action dataset for dexterous manipulation
The research team releases TacClaw-Infinity, a large-scale structured vision-tactile-language-action (VTLA) dataset for contact-intensive manipulation, with a scale exceeding 10,000 hours, 1 million episodes, and 1 billion frames, covering more than 1,000 task types.
arXiv cs.ROSource ↗Any3R open-sourced: 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 and can reconstruct a 3D point cloud and estimate poses 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, enabling a single set of weights to cover pinhole, fisheye beyond 180°, and 360° panoramic cameras. On benchmarks across 13 central camera types and 3 datasets, it leads in both reconstruction accuracy and pose accuracy, and its inference code, 13 camera models, and weights have been open-sourced.
3D视觉工坊 · 微信公众号Source ↗
Brief updates
SimuGuard: Auditing contact artifacts in simulation benchmarks such as LIBERO and RoboTwin
The authors propose SimuGuard, an audit plugin for LIBERO, RoboTwin, RoboCasa, and RoboDojo that identifies failures caused by the simulator rather than the policy. On RoboTwin, the authors trace such failures to the physics engine's collision detection: when geometries are 1 to 2 centimeters apart, the engine reports about 2 centimeters of penetration, and the solver ejects objects at several meters per second, because the distance algorithm discards triangles below a fixed area threshold.
arXiv cs.ROSource ↗∇Mimic: Efficient humanoid whole-body control with differentiable simulation
The authors propose ∇Mimic, which uses exact gradients from a differentiable simulator to train humanoid whole-body motion tracking policies, reaching a score of 0.82 in 20 minutes and 0.874 in 1.8 hours on a single GPU, whereas world-model methods require 1.2 hours to reach 0.82.
arXiv cs.ROSource ↗From teleoperation to long-horizon dexterous manipulation: Evaluating visuomotor skills and offline proxy metrics
The research team used a five-fingered robotic hand to grasp and insert RJ45 connectors, evaluating 281 imitation learning policies covering seven policy families across about 1,700 real-robot trials. Controlled ablations show that observation and action representations have a greater impact than conventional optimizer and sampling choices; among point-cloud policies, region-of-interest cropping, robot-base-frame representation, and normalization within the point encoder each improve success rates by 70 to 100 percentage points, and vision-language-action policies switching to absolute action prediction improve insertion success rates by up to 70 percentage points.
arXiv cs.ROSource ↗