The paper proposes Embodiment-Classifier Guidance, which trains a lightweight embodiment classifier and uses the classifier's gradient during denoising steps to guide a frozen π0 policy, enabling transfer to new robot embodiments without additional demonstrations or parameter updates.
The arXiv paper Micro-Localization demonstrates high-precision global localization using microscopic optical observation of natural ground texture, achieving localization accuracy on the order of tens of micrometers. The authors established the relationship between optical magnification and image-space localization uncertainty on the one hand and physical localization accuracy on the other, and addressed the calibration, camera modeling, and focusing challenges introduced by high-magnification imaging.
An arXiv paper proposes RePlanMem-VLA (Replanning-Guided Keyframe Memory for VLA), which turns a policy's own plan revisions into an online memory-write signal: during receding-horizon execution, if consecutive action chunks show unusually large disagreement in their predictions of the same future action, this is treated as a task-relevant event.
The paper proposes RTISSLAM-3D, a 6-DoF SLAM system for a wheeled robot equipped with an in-air 3D imaging sonar, an IMU, and wheel encoders. The system uses a spherical CFAR detector to reduce sonar images to sparse 3D detections, which are used for 3D acoustic odometry that models the sonar beam shape, for place recognition with polar submap descriptors, and for loop-closure verification via 3D registration, all combined into a 6-DoF pose graph with attitude factors.
The paper proposes RAX (Robot Agent eXploration), an open-source robot agent framework centered on scaling online exploration, in which a general-purpose coding agent generates robot control code without using action demonstration data, hand-engineered action primitives, or value-function guidance.
Why it matters: The paper compares the imitation learning paradigm that relies on demonstration data with a demonstration-free online exploration approach, and reports quantitative gaps in simulation and on real robots, helping readers assess the feasibility of this direction.
The SenseNova joint research team released a report evaluating GPT-6-Astra's embodied capabilities in robotic manipulation and navigation. Under a unified protocol, the team evaluated it on 12 representative benchmarks spanning tabletop manipulation, humanoids, mobile platforms, and dexterous hands, allowing in-context learning (ICL) from training-set demonstrations and substantially increasing the number of execution steps, achieving what the paper calls unprecedented performance.
Mothership proposes a two-tier autonomous refuse collection system in which an electric compaction truck carries two low-speed sidewalk robots, RoboCarts, that fetch and deliver standard two-wheel bins to the truck, return them to their positions, and re-dock for charging. A discrete-event simulation over synthetic campus layouts at 2-16 bins/ha quantified throughput: a two-cart configuration serves 20-29 bins/h, while three to four carts can reach 35-49 bins/h at an 80 m docking radius; the optimal docking radius must be chosen together with fleet size.
SpatialScout is a new benchmark for evaluating vision-language model (VLM) agents' ability to actively build object-level 3D maps. It contains 80 diverse indoor scenes and 3,140 target objects. The task requires agents to control a camera to explore scenes and estimate object centers in a shared coordinate frame, coordinating viewpoint selection, cross-view association, geometric estimation, and map revision within a limited interaction budget.
PhysSkin++ proposes a physics-informed framework that uses a Neural Skinning Fields Autoencoder to learn continuous neural skinning fields from static geometry, constructing a compact deformation subspace without per-object network training or simulation trajectories.
The paper proposes Ropedia W0, an egocentric loco-manipulation world foundation model spanning humans and robots with diverse embodiments. The research builds the Xp-Data dataset, containing 10,000 hours of egocentric human data and 3,000 hours of robot data, with unified multimodal annotations and quality control; camera poses encode locomotion and projected skeletons encode manipulation, with robot skeletons combining each embodiment's forward kinematics and manual keypoint annotations.
FocusGUI proposes a lightweight adapter that achieves focused perception for omni-modal GUI agents by dynamically weighting multimodal channels at each step, compressing key evidence into a compact latent, and injecting it into the frozen decoder's embeddings. Across four omni-modal backbone models, it raises Type Match by an average of 33.2 and Exact Match to 3.0x relative to the frozen backbone, and improves Type Match by 30.1 over the strongest existing enhancement method. Training relies only on action supervision, requiring neither modality-relevance labels nor backbone fine-tuning.
The paper proposes Anti-Persistence, a memory design that accumulates trajectories and outcomes, reconstructs task guidance on demand, and discards it after execution, so that errors in stored abstractions are not repeatedly reused. The authors call the phenomenon in which errors are regenerated through a cycle of abstraction, execution, and re-abstraction error reproduction. Across multiple benchmarks and backbone models, Anti-Persistence achieves higher held-out task success rates and the lowest average LLM call cost among the evaluated memory designs.
Researchers propose PVBench, a physical video benchmark that quantitatively measures the physical plausibility of generated videos across 46 dimensions in seven domains, covering instruction and semantic fidelity, behavioral and causal consistency, perceptual and temporal quality, geometric and spatial consistency, material properties and field states, kinematic and spatiotemporal persistence, and mechanics and conservation laws.
UC Davis and ADI have open-sourced Agile-WAM, a tactile world action model for contact-rich manipulation. Through multi-scale, multimodal prediction, it lets vision be supervised over long spans and touch over next-frame spans, achieving a 29.4% relative improvement in overall success rate across five real-robot tasks. On a single RTX 4090, inference latency is 10.35 ms with a control frequency of 96.62 Hz, about 40 times higher than the diffusion policy DP; it achieves best or near-best results on multiple of 9 ManiFeel simulation tasks and Flexiv Rizon 4 real-robot tasks.
A team at HKUST-GZ proposes UniWAM, which uses a Mixture-of-Transformers (MoT) architecture to integrate physical reasoning, future visual generation, and action prediction into one model. Built on three expert modules—Qwen3-VL-2B-Instruct, the Wan2.2-TI2V-5B backbone, and an action predictor—and paired with the EgoANT pipeline for automatic annotation of egocentric human video, it is jointly pretrained on human data, robot demonstrations, and VQA data.
StochasticNCBF uses stochastic neural control barrier functions to achieve safe control of an inverted pendulum, analyzing the stochastic generator, safety control correction, and second-order neural barrier propagation. Reproduction experiments compare the safety performance of three schemes: no correction, deterministic correction, and retaining the stochastic term.
EgoWAM examines a concrete deployment conflict in dual-arm robotic manipulation: first-person human data collected by Project Aria glasses is cheap and abundant, but retargeting human hand trajectories into robot actions for behavior cloning (BC) is difficult, so this work switches the supervision channel from actions to world state.
VLA-Rolling is a lightweight inference framework that reuses the action tails discarded during rolling planning by flow-matching VLA models. Through three stages—action-tail inheritance, per-action confidence estimation, and renoising with denoising step allocation—it reuses denoising progress while keeping the underlying VLA frozen.
PhysLDM proposes a unified latent diffusion paradigm for one-shot high-fidelity deformable body simulation. At its core is a holistic spatiotemporal VAE that avoids the staircase artifacts of standard temporal compression, achieving approximately 2.48 mm reconstruction precision on meter-scale scenes with up to 78x token compression.
The paper proposes BiDex-100, a simulation benchmark for dual-arm dexterous manipulation comprising 106 tasks organized along five capability axes: generalization, instruction following, limited demonstrations, long horizon, and memory. Each task is paired with an out-of-domain variant featuring controlled visual and spatial changes to assess robustness.
A Perspective article argues that Joint-Embedding Predictive Architectures (JEPA) are moving from representation learning toward latent world models that support prediction, planning, and control, and that interpretability research must shift from "what is encoded" to how latent variables interact, evolve, and causally drive prediction.
AffordanceWAM is an affordance-aware generative world-action model that represents object-centric spatiotemporal affordance through Scalar Affordance and Affordance Heatmap within the generated future world, grounding visual prediction in task-relevant objects and interaction regions.
CoEvo-RSI proposes a recursive self-improvement (RSI) framework that advances the co-evolution of models and execution frameworks through agent-driven bidirectional historical inheritance, using a meta-agent to coordinate framework design with model checkpoint selection or merging to mitigate the "brain-body mismatch" problem.
The paper proposes the HumanoidPose framework, which treats vision as proprioception and recovers a humanoid robot's full-body joint configuration in world coordinates from video alone. The method combines pretrained human pose features with robot kinematics in a unified model and retargets human motions to sixteen different humanoid embodiments, enabling scalable learning entirely from synthetic data; it also uses unlabeled online videos through self-training to develop a 2D robot pose detector to improve real-world generalization.
LearningHumanoidWalking uses a clock-conditioned reinforcement learning method with footstep planning for humanoid walking, and explicitly models actuator back-EMF. The official weights completed a 200-step deployment validation in MuJoCo, with pelvis height remaining stable throughout.
AirCombat uses rule-based agents to simulate UAV air combat swarm engagements ranging from 1v1 to 10v10, covering close-range dogfights and ten-versus-ten scenarios. The system breaks air combat decision-making into 48 discrete actions and introduces an abstract weapon engagement model and a formation target assignment mechanism to explain the sources of tactical decision-making complexity in multi-aircraft confrontations.
asRoBallet achieves dynamic balance control of spherical underactuated robots through friction-aware reinforcement learning. Starting from underactuated dynamics and the two-layer contact friction among wheel, ball, and ground, it derives the kinematic mapping and dynamic balance mechanism of the spherical chassis. The method uses proximal policy optimization (PPO) to train position-holding and velocity-tracking policies, and applies friction domain randomization to improve policy adaptability under different friction conditions.
Running a real ONNX policy in physics simulation, whole-body angular momentum is computed per body for the humanoid robot G1's Double Spin Kick, breaking down spin and revolution, the scheduling of each body group, and landing impact. The analysis notes that after the double spin kick leaves the ground, total angular momentum is locked, and the ground-contact phase is the window for momentum injection and absorption.
A G1 shooter based on motion-prior reinforcement learning scored on all 60 penalty kicks, and a goalkeeper using historical encoding saved 85% of shots. However, when the two models were placed in the official adversarial scenario, they scored zero goals across 20 matches; the article uses frame-by-frame data to identify a two-layer distribution-boundary mechanism.
The University of Hong Kong's MMLab team proposes EgoWild2Dex, which directly trains a dual-arm dexterous robot on 538.9 hours of unscripted human first-person manipulation data, using under 1 hour of real-robot data per downstream task; across three long-horizon tasks—unpacking a parcel, fixing ornaments with a hot glue gun, and making a cold drink—the average success rate is 96.7%.
A team at the Hong Kong University of Science and Technology developed the embodied robotic arm EmArm, embedding a soft optical tactile skin into a rigid skeleton to achieve whole-arm tactile sensing and adaptive manipulation, with tactile sensing fidelity reaching 97.9% under five environmental disturbances. Its tactile skin covers over 71,000 square millimeters with a spatial resolution of 0.085 millimeters, can recognize five touch behaviors and provide emotional feedback, with recognition accuracy up to 99.8%. The research was published in Nature Sensors.
A Zhejiang University team proposed GuardPIBT, which, while retaining the PIBT execution kernel, uses a neural network to make residual reranking proposals over PIBT's native candidate actions; conflict checking, priority inheritance, and backtracking are still handled by the PIBT kernel, and counterfactual grouped gating is introduced to filter harmful interventions.
Scale AI, in partnership with Elorian, released a new benchmark called Humanity's Sixth Sense (HSS) to test intuitive visual reasoning in multimodal models: 20 human participants scored 93.1%, while top-ranked GPT-6-astra reached only 53.6% even at maximum reasoning effort, and the median across 25 models was 30.9%.
The cartpole repository released the 2d-cartpole-3-link-v1 benchmark, whose spatial swing-up and hold policy passed 32/32 fresh noisy 3D starting-state qualification tests in MuJoCo 3.3.5.
The corbensorenson/cartpole 2D Cartpole four-link benchmark passed 32/32 spatial swing-up and hold trials from fresh noisy 3D starting states, with each episode lasting 30 seconds and at least 7.22 seconds upright.
The corbensorenson/cartpole repository released 2d-cartpole-5-link-v1, a five-link 2D spatial swing-up and hold benchmark that passes qualified swing-up in native MuJoCo 3.3.5 with 32/32 fresh noisy 3D initial states.
A two-stage joint optimization calibration method for LiDAR-camera extrinsics assisted by monocular depth is proposed, replacing traditional grayscale-intensity mutual information with depth constraints to improve calibration accuracy and robustness under complex lighting and material conditions. In the coarse calibration stage, a composite calibration board combining ArUco markers and a circular-hole array is used, and an SVD-optimized ICP algorithm solves for initial extrinsic values analytically without requiring preset initial values; in the fine calibration stage, the DepthPro monocular depth estimation network is introduced to construct a joint optimization objective function combining geometric constraints and normalized mutual information distance (NID).
Professor Wang Jingyuan's team at Beihang University, together with the School of Management at Shanghai University, published the paper MoE-Pointer at KDD 2026, defining the Dynamic Multi-Stage Pickup and Delivery Problem (DM-PDP) and using a sequence-to-sequence generation framework to achieve end-to-end reinforcement learning joint optimization of a three-stage rider-drone delivery chain.
WorldOracle officially opened submissions to developers on October 16. It is an evaluation platform that uses real execution records as a reference to test the prediction capabilities of action-driven world models. The paper is titled "WorldOracle."
In the paper "Fingers as Legs," the ETH Zurich research team used reinforcement learning to enable the WUJI humanoid dexterous hand to crawl, turn, and recover from falls using its own fingers, without a separate mobility mechanism.