A project video for the DIY bipedal robot CBR-II reviews the complete process of training walking policies with reinforcement learning and deploying them from simulation to the real robot, and shows the final results on the hardware. The robot can now walk and stand on one leg.
Afferent-Omni is an open-source visuo-tactile command and control (C2) framework for embodied intelligence and dexterous manipulation, offering real-time visuo-tactile perception, closed-loop manipulation, and embodied AI evaluation capabilities. Its FastAPI backend includes 9 modular routers and 15 REST endpoints, and achieves closed-loop operation via a 50 Hz WebSocket stream; it integrates a neuromorphic reflex arc with under 15 ms latency, multi-finger topology, and acoustic friction profiles.
The underwater robot sector saw a concentration of financing and orders in 2026: Shihang Intelligent's orders for the year exceeded 1 billion yuan; Shenhai Zhiren signed 8 work-class ROVs in South America and Southeast Asia with total contracts of 300 million yuan; and Ouka Zhibo won two consecutive water-area cleaning projects totaling about 27.17 million yuan. The article uses Shenhai Zhiren, Weilan Shenhai, and Ouka Zhibo to represent three hurdles—selling equipment, completing tasks, and replicating platforms—and notes that scarce underwater data, unproven stable profits, and a shortage of onshore engineering talent remain the parts of the industry that have not been settled.
General-purpose large models are beginning to take part in robot control. Models such as GPT-6 Astra can already understand complex instructions, recognize objects, and plan operation steps in tests, but fine manipulation still requires coordination with dedicated control systems. Approaches such as HexaShell and WorldScape-RSI explore a division of labor of "large-model decision-making plus specialized-model execution," and tests show some improvement in task performance, though the results mainly come from controlled environments. On data collection, head-mounted operation videos are quoted at about 100 to 300 yuan per hour, while complex real-machine data can reach about 1,000 yuan per hour.
Schaeffler has opened an approximately 7,500-square-meter humanoid robot lighthouse factory in Taicang, China, where six humanoid robots at five workstations handle the loading, sorting, assembly, inspection, labeling, and packaging of key components such as linear and rotary actuators. The factory also includes a training and data acquisition center, a digital twin platform, and a data center, enabling real-time mirroring and optimization of parameters such as assembly precision, work hours, and energy consumption, and offering customers and partners the ability to train and test physical AI systems under real industrial conditions.
Ranked by growth metrics such as revenue CAGR and CAGR of non-recurring net profit, and profitability metrics such as return on equity and gross margin, the article selects 10 companies in the robot data collection industry chain. Longcheer Technology, Orbbec, and OPT Machine Vision take the top three spots. The list covers segments including embodied data collection hardware ODM, RGB-D depth collection hardware, optical motion capture, professional embodied data services, and data collection center operations. Frontop Digital Creative Technology's revenue CAGR of 72.81% is the highest in the sample. The data is based on historical figures and does not constitute investment advice.
The authors propose TouchSafeBench, a physically grounded benchmark built in Habitat 3.0 for evaluating the collision grounding capability of vision-language models (VLMs). It contains 2,940 simulated indoor co-presence episodes covering two scenario types, social navigation and social rearrangement, and provides synchronized multi-view RGB-D observations, top-down trajectory maps, calibrated camera metadata, and simulator contact labels.
The authors propose TopoGS-SLAM, which decouples refinement from capacity using coverage-triggered Gaussian initialization and bounded-topology optimization, turning mapping into a bounded-dimensional problem. On standard indoor benchmarks, its localization and reconstruction accuracy is not inferior to densification-based pipelines, while memory scalability and throughput improve significantly. The authors therefore conclude that in bounded-area static indoor scenes with sufficient multi-view coverage, Gaussian SLAM does not require iterative topology mutation.
The authors propose ContactBench 1.0, which uses 36 reference tasks and more than 40 object assets to break contact-rich manipulation down into eight essential skills; the assets can be reproduced on consumer-grade 3D printers and each comes with a matching simulation model. They evaluate recent vision-only and visual-tactile policy learning methods on six benchmark tasks, and the results show that vision-only policies struggle to perceive or respond to contact events, while visual-tactile policies do not bring noticeable improvement. The benchmark will be open-sourced, providing physical and simulated assets and controlled task variations for evaluating the generalization ability of robotic systems.
ArtifactArena is an open-ended evaluation platform that challenges models with hardware-software co-design: building working robots in a simulated arena and pitting them against each other. The platform guides models to improve their robot artifacts through three channels: text descriptions, physics simulator feedback, and gameplay data, and compares frontier models using Elo rankings from head-to-head tournaments. The authors are releasing the framework and tournament infrastructure for ongoing community submissions, creating a non-saturating long-term testbed.
The authors propose WorldEcho, which evaluates whether action-conditioned world models truly follow commanded actions over a broader action distribution through visual integrity and SE(3) end-effector trajectory alignment. The diagnosis shows that models trained on expert data perform well on demonstrated actions but produce action-inconsistent or visually invalid generations on non-expert trajectories.
This paper has been withdrawn by its author, Zeyu Ni. OccamView is an object-conditioned view-selection framework for frame-budgeted active 3D Gaussian reconstruction. It builds an online object memory from open-vocabulary detections, represents unresolved local occupancy around detected objects as conservative hidden-region proxies, and evaluates candidate viewpoints with an occlusion-aware proxy-coverage score.
The paper proposes Semantic Anchoring, a plug-and-play method that anchors VLA action representations to a semantic manifold and splits them into a shared semantic channel and a private channel; both are discarded at inference, leaving the deployed model unchanged.
This survey examines three roles of 3D generation in embodied AI and robotic simulation: as a data generator producing simulation-ready assets with articulated structure, physical properties, and deformable content; as a builder of interactive, task-oriented scenes for simulation environments; and as a sim-to-real bridge supporting digital twin reconstruction, data augmentation, and synthetic demonstrations. The authors note that the field is shifting from visual realism toward interactive usability, and they list key bottlenecks including limited physical annotation, the disconnect between geometric quality and physical validity, fragmented evaluation, and the sim-to-real gap.
The authors propose VORL-EXPLORE, a hybrid learning and planning framework that uses execution fidelity to estimate local navigability and execution reliability, coupling it into a Voronoi allocation objective with inter-robot repulsion and arbitrating during navigation between global A* guidance and a reactive reinforcement learning policy. The fidelity model is recalibrated using self-supervised pseudo-labels generated from recent progress and safety outcomes. Experiments on randomized dynamic grids and an ablation under severe congestion show higher success rates, shorter exploration paths, and lower inter-robot coverage overlap, along with a deployment example in a Gazebo factory scenario; the code is open-sourced.
A research team completed an on-orbit demonstration on the International Space Station's Astrobee free-flying robot of machine learning-based warm starts to accelerate trajectory optimization, using a neural network to predict initial values for a sequential convex programming (SCP) solver, which then enforces safety constraints. In scenarios including rotational dynamics, the trained network reduced the number of solver iterations required for convergence by 60%, and by 50% in obstacle scenarios drawn from the training distribution. The work was published at iSpaRo 2025.
SceneBloom proposes an agentic framework that combines visual layout sampling with contract-guided 3D compilation to generate realistic and diverse simulated indoor environments. Its hierarchical sampler uses diffusion priors to generate furniture layouts within functional zones, and the compiler then performs geometric fitting and local pose correction based on relations extracted from each proposal.
RobustRLlib is an algorithm-centric library and benchmark for robust reinforcement learning that integrates 17 representative online, offline, and safe reinforcement learning algorithms under a unified interface. Its evaluation platform covers a variety of real-world shifts as well as humanoid robot and vision-language-action tasks, and it presents the capability distribution of current robust reinforcement learning algorithms across 273 perturbation configurations, identifying which scenarios remain effective and where substantial robustness gaps persist. The project provides open-source code, a project page, and tutorials to facilitate reproduction and extension.
The authors propose SMM-AC, a regularized reinforcement learning algorithm based on the soft mellowmax value function. On MuJoCo continuous control benchmarks it is comparable to baselines such as SAC, and on real robots it achieves higher returns while being more robust to perturbations in observations, the environment, and dynamics. The study compares the sim-to-real transfer performance of commonly used reinforcement learning algorithms and finds that regularization improves policy robustness to the sim-to-real gap and to real-world environment changes.
The study proposes learning a speed prior from large-scale demonstration data: given a task instruction, a motion path, and images along the path, it predicts a distribution of execution speed along the path and uses a selected quantile as a unified clock to retime each demonstration. The method fits the speed prior on robot teleoperation data and on human handheld-gripper demonstrations, retimes robot demonstrations to the handheld pace, and then fine-tunes a vision-language-action policy; at execution, a real-time optimal-control layer tracks the predicted action chunks under robot constraints.
The authors propose the SAGE (Semantic-to-Action Grounded Execution) framework, which uses trajectories output by a vision-language model (VLM) as preferences over complete geometric candidate paths, and combines each candidate's geometric score with its agreement with the VLM trajectory through rank fusion.
The StackGen3D framework generates stable block assembly structures from a given block inventory and three orthographic silhouette images. It first uses a physics-informed graph neural network (PI-GNN) to model support relationships between blocks and to curate a dataset of stable structures, then trains a conditional flow-matching model to learn the multimodal distribution of block configurations that match the target silhouettes. In preliminary experiments, Best-of-10 selection raised physical stability from 45.3% to 72.4%, with a silhouette IoU of 0.970.
Using matched simulator and model responses, the study measures LeWorldModel's (LeWM) sensitivity to small action perturbations via Gram matrices across nearly a thousand states in Reacher and OGBench Cube.
DexPort is a hand-held data collection interface for high-DoF robot hands that uses the same hand hardware for both collection and deployment, recording joint target commands and measured joint states. The operator moves the hand-held interface to specify wrist motion and coordinates finger motion by selecting or adjusting skills, without needing to control each joint individually.
ViT-H proposes a generation-plus-tracking video-to-humanoid control framework. It uses a diffusion model to generate humanoid reference motions directly from monocular video, skipping human motion reconstruction and retargeting, and uses motion-tracking results to post-train the generator with direct preference optimization. The authors also present an autoregressive variant that enables real-time humanoid video teleoperation with sub-400 ms latency. Experiments on the Unitree G1 show that ViT-H outperforms existing reconstruction-retargeting-tracking pipelines in motion quality and executability, and that performance improves with model size and with pre-training and post-training data scale.
RynnUnify is a 16B unified embodied model that combines embodied understanding with long-horizon visual planning through interleaved video-language reasoning, covering spatial reasoning, physical prediction, and grounding. The authors also developed an agentic extension based on reinforcement learning to selectively use video prediction to reason about unobserved states and interaction outcomes, and a world-action extension that translates understanding and prediction capabilities into robot control.
HoloBoxing proposes a humanoid robot boxing learning approach built on a pretrained whole-body controller, separating reusable motor skills from task decision-making: a compact planner outputs future body-point targets, while the frozen controller handles joint-level execution.
The study proposes an action-aware quantification method that uses a vision-language model to decompose long-horizon manipulation trajectories into object-level subtasks and assign weighted action credits, and builds a training video dataset of more than 800 hours plus the LOOM benchmark, which contains 19,618 human-calibrated evaluation frames.
The authors propose VERID, a persistent, high-fidelity semantic digital twin for automated laboratories that provides AI agents with four core services: queryable introspection, time-compressed simulation, comprehensive trajectory persistence, and a safe execution framework. Three case studies using mixtures of water-soluble dye and deionized water validated its optical and fluidic foundations, demonstrating pre-detection of critical protocol errors, in-run stateful error recovery, and post-hoc identification of subtle physical faults.
UAV-SwarmAgent is a hierarchical multi-agent system in which an operator issues a mission in one natural-language sentence, and a ground-station agent generates and executes code that turns it into routes for each UAV. The system uses a vision-language model to locate inspection targets on an aerial map, balances them among N UAVs by route length, and then worker agents equipped with lightweight decision models verify incidents.
The authors propose HGWM, which combines an LLM-generated room-anchor prior with persistent spatial-semantic memory to select search directions in zero-shot object navigation. On HM3D v0.1, it reports 62.6% success rate and 31.5% SPL, versus 58.1% and 31.2% for WMNav; on MP3D, 46.8% and 17.9%, versus 45.4% and 17.2%.
The study proposes an anatomy-anchored imitation learning framework that combines anatomy-guided probe placement with learned ultrasound navigation. Across 16 unseen participants, task success rates for standard-view acquisition of the kidney, gallbladder, and liver and for radial artery scanning were 96%, 94%, 88%, and 94%, respectively.
CoCarry introduces a dataset and a contact-support-structured retargeting method for dual-humanoid cooperative carrying. The dataset contains 220 synchronized dual-person sequences spanning 21 task schemes, 5 shared-object geometries, and 4 terrain types, totaling about 61 minutes of motion data.
LogLook proposes an execution-log-style memory that lets frontier vision-language model planners record executed actions and earlier notes, and selectively review image frames based on expected changes. With the benchmark's built-in motion planner, it matches the success rate of note-only memory while using 43% fewer tokens and less than half the images, and it causes notably fewer failures due to memory or observation errors; under direct control, it solves 93.5% of the RoboMME test set.
DynaSlots is an action-label-free 3D visual pretraining framework that decomposes each point cloud frame into a small set of temporally consistent slots and models dynamics at the slot level rather than the scene level. A temporal slot propagation module keeps slot assignments consistent across frames; an inverse dynamics model extracts a latent action for each slot from feature differences; and a diffusion-based forward dynamics model, paired with a self-supervised interaction head, predicts each slot's future state and learns pairwise proximity and motion relationships.
MAST-VLA proposes a dual-conditioning framework that introduces skill-conditioned LoRA Mixture-of-Experts adapters into the LLM backbone and converts predicted affordance heatmaps into a patch-level trajectory-affordance spatial prior for the action head. In the CALVIN ABC→D long-horizon multi-skill cross-scene setting, it achieves an average completion length of 4.65 over five-subtask sequences, the best performance in that setting. Real-robot multi-skill sequence experiments further validate its deployability under real perception and control noise. The paper has been accepted to CoRL 2026.
The study tested motion-tracking policies for a 24-joint humanoid in an independent CPU MuJoCo environment and found that a constant 40 ms command delay reduced the average survival rate from at least 26/30 under training randomization to 15%, with a delay margin of only 20–30 ms.
The research team deployed a VLA model on a Unitree Go2 quadruped equipped with an AgileX Piper robotic arm, achieving closed-loop mobile manipulation, and reported choices such as model variant, action horizon, and demonstration conventions that made real-robot deployment feasible. For the pick-and-place task, expanding demonstration coverage reduced held-out end-effector error by 30%, and adding secondary skills such as pushing introduced new behavior modes without degrading baseline performance. Closed-loop real-robot experiments showed that dataset composition affects task success.
RAISE-Vision is an automated platform for contact angle (CA) measurement on curved hydrogel surfaces, intended for in vitro wettability analysis of contact lenses. Its 3D-printed lens mounts, camera holders, and drying apparatus enable an automated, repeated sessile-drop workflow within a liquid handling system. A custom open-source image processing script combines elliptical fitting and the tangent method to handle droplet deformation on curved surfaces. In 10 consecutive measurements of a senofilcon A lens in air, the contact angle rose from 62.6±5.5° to 83.4±2.1°, with no significant difference from the manual protocol results (p=0.470).
TravSUITE is a traversability system for high-speed off-road navigation, consisting of a voxel mapper that uses visual foundation models to build a geometric-semantic local map and a unified neural network that jointly predicts geometry, semantics, speed, and cost in bird's-eye view (BEV).