HC2Robot proposes a deployable perceptive humanoid parkour framework composed of a Predicted Foresight Module (PFM) and a Hyper-Connection (HC) Transformer. The PFM predicts, from deployable observations, the timing, location, and support quality of each foot's next touchdown event, converting privileged supervision into a deployable latent representation.
TAPESTRY treats robot actions as single-patch tokens in the latent space of a video model, adapting a pretrained video diffusion Transformer into a robot policy that requires no video generation at inference time.
For the high-school remote-control division of the 2027 NASA Human Exploration Rover Challenge, a study presents an integrated, reproducible design assessment of a four-wheel educational rover: 10.919 kg, 215 mm wheels at 200 RPM give a theoretical rolling speed of 8.105 km/h, and the nominal motor specifications are insufficient to demonstrate compliance with the 8 km/h design ceiling.
An arXiv paper proposes modeling bio-inspired magnetic navigation (BIMN) as a gradient flow control and guidance problem, using magnetic inclination and intensity of Earth's magnetic field to guide a system through intermediate waypoints to a destination within a predefined region. The authors note that magnetic coordinates do not map one-to-one to geographic locations and are ambiguous, but that gradient flow can act as a "water current" carrying the system's motion and be steered. The framework could be used in engineering to build geomagnetic navigation infrastructure and to define and follow prescribed routes, and also to analyze animal migration routes that rely on magnetoreception.
RynnWorld-Latent is a cross-embodiment video world model that uses the 3D-grounded latent actions introduced by RynnLAM as a shared conditioning interface, enabling joint pretraining on human interaction videos, robot demonstrations, and simulation data without requiring a unified native action space.
The research team proposed Torch-Hand, a humanoid dexterous hand featuring an all-linkage transmission architecture with 21 degrees of freedom, including 5 DOF in the thumb and 4 DOF in each of the other four fingers. The hand integrates its actuators into the palm in a serial-robot-like arrangement, so that motor motion corresponds directly to joint kinematics while balancing grasp force and payload capacity. Experiments evaluated the reachable workspace, fingertip coordinated trajectories, thumb opposition, coverage of 33 grasp categories, grasp force, payload, and task-level manipulation.
The authors propose a traceable framework comprising four components—LAM, OpenSDL, Safe-SDL, and L3-Bench—to connect experiment planning, automated execution, and measurement feedback. LAM is a measurement-conditioned policy architecture that extends the vision-language-action paradigm to generate both continuous robot trajectories and typed instrument API calls; OpenSDL maps role-level operations to heterogeneous devices and records timestamped causal events.
The paper proposes a method for enabling two personal mobility vehicles (PMVs) to drive autonomously side by side: after a global path is given, each vehicle runs its own sampling-based model-predictive planner, the two exchange only the odometry information each can observe of the other, and a common rule is used to make a joint decision on a utility matrix over pairs of candidate trajectories.
Using a sanding task as a case study, the paper combines simulation with experiments on two lightweight, unactuated wheeled testbeds to characterize non-rigid base effects in mobile manipulators. Under low base-ground resistance, only 54-70% of a 100 N command was realized, with tool-center-point errors reaching 100 mm; increasing base-ground resistance during contact improved force realization and arrested further tool displacement, while base vibration and force-tracking error showed no consistent relationship under the tested conditions. The study was conducted by Zongyuan Zhang et al. and has been accepted to ACRA 2026.
TOPS proposes a task-oriented prior and scheduling method that adapts the flow-matching source prior and interpolation schedule for each task without modifying the network architecture. On LIBERO-10, single-step inference raises the success rate from 53.44% for standard flow matching with ten steps to 69.67%, with inference time about one-sixth of the baseline. The method improves performance on LIBERO, LIBERO-Plus, and CALVIN, and when transferred to the π0.5 policy, and on a real robot it raises the success rate on an upper-drawer-opening task from 20% to 80%.
RoboSTD is a zero-shot single-to-dual data construction method for bimanual robot learning. It applies sagittal-plane mirroring to observations, proprioception, and actions to generate executable contralateral pseudo-demonstrations, then uses an LLM to generate structured spatio-temporal coordination constraints that guide the reconstruction of pseudo-bimanual supervision across workspace regions and task stages. The study evaluates RoboSTD on the AgileX ALOHA platform in RoboTwin simulation and real-world settings. Without additional bimanual demonstrations, it achieves an average of 2.11 times the number of cups collected by the single-arm baseline in the cup-collection task.
The paper proposes a learning-based real-to-sim-to-real framework that converts biological curves into cable-driven modular soft robot configurations through surrogate-assisted inverse design and uncertainty-guided refinement. The learned forward surrogate reduces deformation prediction time from 1.53×10^5 ms to 0.0008 ms, a speedup of about 1.9×10^8 times; inverse design time drops from 1.99×10^6 ms to 4.18 ms.
SWIM proposes using an inverse dynamics model to learn state feature weights from offline trajectories for state-supervised training of JEPA world models. The method predicts recorded actions from current states and observed future displacements, quantifies each state feature's contribution to action prediction, and uses this to build a weighted state metric that aligns the latent space with state distances. The inverse dynamics model is used only during training; planning still relies solely on images. On five simulated robotic manipulation tasks, SWIM's planning success rate consistently outperforms uniform state weighting.
The authors propose HumiTransfer, a framework that converts in-the-wild human-human interaction videos into physically feasible human-humanoid interaction motions. The method first reconstructs 3D human motion with foot-ground contact, then transfers one participant to the target humanoid embodiment through interaction modeling, and uses post-retargeting optimization to ensure balance, contact, and collision avoidance.
The authors propose QuadPose, which estimates the full-body 18-DoF pose of a quadruped robot from monocular video sequences, including the 6-DoF base pose and 12-DoF joint angles, and release QuadSyn, the first camera-to-quadruped robot pose benchmark. On QuadSyn, QuadPose achieves 92.18% PCK@2.5px and 85.94% ADD@20cm, with a mean joint-angle error of 3.65 degrees and a runtime of about 33 FPS.
OpenJaws is a fully open-source sensorized UMI hardware and software framework that pairs a modular 3D-printed gripper with a low-cost stereo camera to recover metric camera trajectories, improving action-label localization accuracy in handheld demonstrations. The authors also propose UMI-Loc-Bench-v1, a benchmark for quantifying the real-world localization accuracy of handheld data-collection devices, and show that stereo localization clearly outperforms monocular alternatives. Policies trained on data collected with OpenJaws have been deployed on real robots across multiple tasks, and the hardware, software, and data are released under a commercially friendly license.
REACH is an RL-based hierarchical framework that enables a Unitree G1 humanoid robot to traverse height-constrained environments through variable-height bipedal walking and crawling. Its policies regulate body height over complementary command ranges: 1.27–0.50m for bipedal walking and 0.50–0.20m for crawling, and a bidirectional transition policy connects crouched walking and crawling. The framework also combines a LiDAR planner that converts point clouds into collision-free three-dimensional paths and locomotion commands. Real-world experiments validated height and speed tracking, terrain traversal, mode switching, and confined-space traversal; the code and videos have been released.
CARD is a simulator- and object-model-free framework that retargets human demonstrations into robot actions by explicitly preserving contact force targets, and it handles video, motion capture, and tactile glove data in a unified way. On video, glove, and motion-capture data and the GRAB dataset, CARD maintains contact on 43%–78% of contact frames, whereas purely kinematic and object-aware retargeting methods stay below 10% and 22.5%, respectively.
The study proposes CPG-ENN, a quadruped locomotion framework that uses a symmetry-based eight-neuron central pattern generator (CPG) to supply gait rhythms, while a pretrained execution neural network (ENN) shared across legs maps rhythmic signals and trajectory parameters to joint-angle references without online inverse kinematics.
The study proposes the MyoAcross framework, which combines individual musculoskeletal model construction, morphology-conditioned policy learning, and virtual model augmentation to achieve muscle-driven reference-motion tracking without per-individual fine-tuning. Across 31 held-out individual models, it achieved a 99.24% completion rate and reduced tracking error by 22.60% relative to an unconditioned multilayer perceptron. Policy reuse also extended to 22 held-out individual models constructed from single-image shape estimates, height, and body mass, as well as to a model with unilateral lower-limb shortening after asymmetry augmentation.
RoboAuto proposes a robot program search framework that organizes physical understanding at three levels—object, scene, and interaction—to guide LLM agents in generating and revising manipulation programs. The method achieves better performance across five benchmark suites, reaching an 89.1% success rate on short-horizon LIBERO-PRO tasks and averaging more than twice the success rate of ASPIRE on LIBERO-PRO Long and LIBERO-Plus.
GRAFT proposes a geometry-aware residual adaptation framework that, without changing the parameters of a frozen humanoid behavior foundation model (BFM) or its motion-reference interface, uses a lightweight adapter to apply terrain-conditioned corrections only to lower-body execution targets. The adapter is trained with task and foot-geometry rewards, is regularized so it produces no unnecessary corrections on flat ground, and is distilled from privileged terrain observations to onboard depth.
SVIn+ is a multi-sensor state estimation framework for autonomous underwater vehicles (AUVs). Building on the open-source visual-inertial SLAM system SVIn, it integrates water-pressure and DVL measurements into tightly coupled nonlinear optimization and uses a KISS-ICP decoupled front end to generate relative pose factors, incorporating 3D sonar structural information into the factor graph. The authors conducted field deployments in shipwreck and coral reef scenarios, validating drift suppression and global consistency. The paper was accepted to IROS 2026, and the code and datasets have been released on the project website.
This study proposes using a stochastic motion tokenizer to generate concise discrete motion representations that can be fed directly into a whole-body tracking policy to execute dance, fighting, and walking motions on the Unitree G1. At matched capacity, the tokenizer reconstructs Unitree G1 motions better than FSQ and VQ-VAE. In real-robot demonstrations, each generated motion lasted 16.8 seconds, and a 35-second fighting-to-dance sequence was completed.
For multiple observations under image-quality constraints in drone-based 3D reconstruction and inspection, where overlap-based planning heuristics cannot guarantee such observations, this work models GSD-constrained k-coverage viewpoint selection as a set multi-cover problem. Candidate viewpoints are generated along outward surface normals, screened by distance, viewing angle, field of view, and occlusion, and then a dynamically weighted greedy heuristic prioritizes coverage gaps and scarce viewing angles. Across two structures, four voxel resolutions, and three coverage requirements, the method achieves full or near-full k-coverage, using only 3.00%–16.39% more viewpoints than ILP.
The authors propose HIDRA, a hierarchical dual-attention routing framework for replay-free, single-pass lifelong imitation learning that decouples instruction-level expert retrieval from context-dependent refinement. The method introduces key-level geometric regularization that constrains within-task alignment and between-task separation to mitigate routing interference under representation drift. On several LIBERO benchmark suites, including variants with heterogeneous task sequences and paraphrased language instructions, HIDRA outperforms replay-free and replay baselines on AUC and forward transfer while maintaining low forgetting, and the code is publicly available.
The authors propose SG^2SLAM, a 3D Gaussian Splatting SLAM method for monocular outdoor navigation that replaces per-frame cross-dimensional correspondence with a chunk-level metric prior, so that tracking and mapping share the same geometric source. The method also introduces a ray-conditioned 3DGS kernel that extends each primitive with a viewing-direction subspace, and jointly optimizes spatial geometry and poses. On the KITTI and Waymo Open datasets, SG^2SLAM outperforms per-frame association baselines in the monocular outdoor navigation task. Code will be made publicly available upon acceptance.
ASPEN is an event-triggered scheduler based on command persistence prediction, designed to reduce repeated semantic reasoning calls in navigation VLA policies. On frozen NaVILA in the cross-dataset RxR-CE val-unseen evaluation, ASPEN reduces VLA calls by 9.5% and estimated inference time by 8.1% (including scheduling overhead), while preserving action sequences identical to the baseline in 92.1% of episodes.
The study proposes extending PPO into phase-conditioned policy optimization, using a phase-conditioned critic, phase-gated rewards, and a multi-phase reset curriculum to train a single policy for long-horizon loco-manipulation tasks. The method adopts a decoupled hierarchical control architecture that separates low-level balancing dynamics from high-level phase execution, and validates multi-stage skills on real hardware on the Unitree G1 and a Boston Dynamics Spot equipped with a manipulator arm, while testing design choices through ablation experiments.
cuArms is a conflict-based planner co-designed with GPU parallelism. It decomposes the search space using reusable multi-modal roadmaps and resolves conflicts and deadlocks through lazy search. It integrates with a high-level scheduler to jointly optimize task assignment, scheduling, motion, and manipulation skills. Experiments report real-time initial solutions and anytime improvements in motion quality and scheduling, without domain training for learning-based methods. The authors say it provides demo videos on both simulated and real industrial tasks.
The paper proposes a program-level adversarial co-adaptive self-play framework that lets coding agents generate attacks and defenses based on opponent code and task feedback, without model fine-tuning. The framework provides external program interfaces for autonomous driving and SDK-level defense interfaces for humanoid robot locomotion; two eight-generation case studies show that the driving defender introduced direct-actuation recovery, while the humanoid defender revised incoming-state validation. These responses provided only local task recovery and did not establish sustained robustness.
Rollback proposes a policy-improvement paradigm in which a world model rewinds to critical decision points, branches to try different actions, and compares the outcomes, distilling counterfactual results into action-level supervision without a separately learned value model. On three real-world tasks—block extraction, ball throwing, and tabletop curling—success rates improved by 10.67%, 11.11%, and 11.85% in absolute terms over the strongest baseline, and ball throwing and tabletop curling surpassed the average human player success rate within 8 hours of training.
The Era AI Lab team proposes E0.5, a geometry-native world-action model built directly on 3D foundation models that treats 3D geometry as the native representation for world modeling and action prediction, and explicitly predicts future 3D states and actions.
The study proposes the Just Action Model (JAM) framework, which breaks down the differences between VLA and WAM into three separable choices: backbone initialization, the future-observation objective, and whether that loss mechanism is retained at inference. It uses a pretrained visual backbone plus an action expert, trained with an action loss and a foresight loss.
The three major A-share indices rose collectively at the midday break. The Shanghai Composite Index rose 0.72%, the Shenzhen Component Index rose 1.99%, and the ChiNext Index rose 3%. The humanoid robotics, communications equipment, and semiconductor sectors strengthened: JPT rose more than 15%, Unitree Robotics rose more than 12%, and 3PEAK rose more than 14%. The liquor, shipping, and telecom sectors led the declines: Nanjing Tanker and Wuliangye fell more than 5%, and China Telecom fell more than 2%.
The Stardust OS (AOS) robot hackathon finals, hosted by Stardust Intelligence, concluded in Shenzhen on October 25. 17 teams developed 16 apps in 36 hours, with 94% running on real robots. Participants built applications by calling AOS-encapsulated Meta capability packages and Skills, combined with large models such as GPT and DeepSeek. The champion team, "Remote Avatar," achieved two-way real-time control between a phone and the T1, completing tasks such as scooping cat litter, watering plants, and picking up packages. Stardust Intelligence founder Lai Jie said that when designing the interfaces, AOS considered both human and AI model users.
Estun disclosed its 2026 A-share restricted stock incentive plan on September 30, proposing to grant 2.99 million shares at a grant price of 14.92 yuan, about 50% of the September 30 closing price of 28.22 yuan. The performance targets are non-recurring net profit of no less than 200 million yuan in 2026 and a cumulative total of no less than 1.1 billion yuan from 2026 to 2028, while the company's historical annual peak non-recurring net profit is less than 100 million yuan.
Humanoid robot output exceeded 40,000 units in the first half of 2026, with full-year projections surpassing 100,000; their performance ceiling is being determined by three foundational material categories: lightweight structural materials, flexible electronic skin, and high-performance batteries. On lightweighting, Tesla Optimus Gen2 makes extensive use of PEEK, which can reduce part weight to about half that of aluminum alloy; electronic skin faces a bottleneck in which high-quality tactile data worldwide totals only about 500,000 hours, a shortfall of over 95%. Current full-size humanoid robots have endurance of only 2 to 4 hours, and most battery capacities are below 2 kWh; solid-state batteries are accelerating from the lab toward robot applications.
ThinkWorld is a VLA learning framework that represents each action thought as a Chain-of-Dynamics (CoD), a sequence of discrete motion tokens that captures the intermediate dynamics from the current world to a future state and serves as a common action representation across embodiments.
Agibot has established Agibot Embodied (Zhengzhou) Intelligent Technology Co., Ltd. in Zhengzhou, with Deng Taihua as legal representative and registered capital of 1 million RMB. The company is wholly owned by Agibot Innovation (Shanghai) Technology Co., Ltd., and its business scope includes intelligent robot R&D, service consumer robot sales, intelligent robot sales, and AI application software development.