The Boreas Road Trip (Boreas-RT) dataset has been released, containing 9 real-world routes and 60 sequences totaling 643 km of driving data. Each route is traversed multiple times, enabling comparison of different traffic and weather conditions in identical environments.
The authors propose the Just Action Model (JAM) framework, which breaks down the differences between VLA and WAM into three separable choices: backbone initialization, the loss objective on future observations, and whether that loss mechanism is retained at inference. JAM unifies both into a structure of a pretrained visual backbone plus an action expert, trained with an action loss and a foresight loss.
The authors propose Constraint-Referenced Strategy (CRS), which projects the end-effector twist commanded by a policy onto the object's joint axis, separating progress along the joint direction from off-constraint waste. On 74 simulated articulated objects, binary success rate only shows a lower transfer failure rate without attribution, whereas the two CRS channels give each a distinct signature.
The paper proposes a denoised counterfactual protocol that uses bitwise-reproducible branching, repeated continuations, held-out oracles, and common random numbers to measure the selectable space a world action model (WAM) can achieve when selecting candidate actions at test time, and builds a fail-gated selector on that basis.
Researchers propose SO-101 Bench, a language-conditioned spatial manipulation simulation benchmark built on an Isaac Lab digital twin of a real SO-101 setup, comprising four task types, 138 objects, 916 evaluation episodes, and 3,629 simulated demonstrations.
AGIBOT's GO-3 Nano Preview topped the latest RoboDojo embodied intelligence benchmark leaderboard with an average score of 61.90, with another result of 56.42%. The leaderboard was disclosed on October 27; specific evaluation conditions and task details were not elaborated in the original text.
We're excited to announce that the Qrevo Edge 3 Pro has been ranked among the top 20 robot vacuums by the pros at @vacuumwars! What makes it one of their favorite Roborock products to date? Watch the full review to find out: https://bit.ly/4ik0jmC
EmbodiedLab is a benchmark that requires robots to "acquire knowledge through action," covering 30 quantitative science experiments across 5 disciplines on 14 apparatus platforms. The parameters of each instance are resampled, so results cannot be known in advance but can be scored precisely.
The latest leaderboard from embodied intelligence benchmark RoboDojo shows AGIBOT's GO-3 Nano Preview taking first place with an average score of 61.90 points and an average success rate of 56.42%, surpassing GPT-6 Astra. In the three sub-dimensions of fine manipulation, long-horizon execution, and open semantic understanding, GO-3 Nano Preview scored 78.65, 68.21, and 61.71 points respectively, ranking first in all three.
The authors propose JRDB-AVR, an active visual reasoning benchmark built on real JRDB robotics data: after receiving a visual reasoning question, an embodied agent must request limited observations by timestamp and viewing angle, and evaluation examines both the final answer and the visual evidence supporting it.
H2RBench is a Real2Sim benchmark for evaluating human-to-robot (H2R) transfer methods. It establishes a standardized protocol based on real human video demonstrations and simulated robot demonstrations, covering four categories of manipulation tasks.
WorldArena 2.0 has been released, expanding embodied world model evaluation along three dimensions: modality, functionality, and platform. In modality, it extends from vision-only to visuotactile; in functionality, it extends beyond policy evaluation and planning to using world models as interactive reinforcement learning environments for policy optimization; in platform, it extends from simulation-only to multiple simulators and real robot embodiments. The benchmark evaluates perception quality, interaction utility, and cross-platform performance under a unified protocol, and the project page provides code and demos.
The authors propose IndustryNav, a simulation environment with 24 warehouse scenes for closed-loop navigation to specified destinations, and introduce safety metrics that record collisions and near-collision warnings. Experiments with 11 leading open- and closed-source VLLMs and 3 navigation baselines show that agents can reach their goals yet collide with moving obstacles, exposing a gap between task success and safe navigation.
The study tested how text placed beside frames affects vision-language models judging robot videos: when a line saying "making progress" was added to clips where the robot was falling behind, Gemini, Qwen, and GPT judges flipped correct verdicts to "progress" 10 to 35 percentage points more often than with neutral text.
This report evaluates GPT-6-Astra as a robot agent on EBench, using the same visual and proprioceptive observation protocol as post-trained policies, and compares two configurations: direct control and harness with policy.
Specola is a tool for evaluating robot behavior in simulation against Signal Temporal Logic (STL) specifications. It provides a Python API and command-line interface that can be integrated into existing policy execution loops. It offers an adapter interface and already provides adapters for nine simulation environments and benchmarks, including LIBERO, RoboCasa, Isaac Lab, and RoboDojo, to supplement success rate with information such as degree of satisfaction, parameter sensitivity, and violated requirements.
EmbodiedLab is a benchmark that requires robots to "act to know": the agent receives an objective and experimental apparatus, chooses its own experimental procedure, reads instruments via camera in a physics simulation laboratory, and is evaluated on the knowledge it ultimately gains as well as the physical state required by the task.
The authors propose Labference, a benchmark for evaluating how robots adjust subsequent operations based on actual physical execution outcomes, separately measuring operation selection, physical execution, and task completion. Preliminary evaluation shows that the tested VLA policies still made target-selection errors even when object transfer succeeded, while GPT-driven control, though able to select correctly, encountered execution failures and higher computational cost.
The study compares cascade PID, hover LQR, BL-SMC, linear MPC with LQR fallback (LMPC-F), a preview-enabled LMPC-F, and a geometric SO(3) controller on a nonlinear plant with rotor gyroscopic coupling, motor lag, and saturation. It finds that a unified twelve-candidate tuning protocol changed the mean-RMSE ranking obtained under fixed baseline settings, moving BL-SMC from last to first; the new ranking recurred on 50 fresh realizations and remained unchanged under a 48-candidate budget.
VLA-HWBench breaks down inference into seven stages for 18 open-source VLA models on NVIDIA Jetson AGX Thor and AMD Ryzen AI Max+ PRO 395, finding that action-generation strategy determines latency more than model size does, with parameter count explaining at most 15% of latency variance.
The authors propose RobotAgentBench (RAB), a benchmark for evaluating robot agents in realistic open-world household environments. Using an agent-assisted, human-in-the-loop real-to-sim pipeline, it constructs 50 household layouts and includes 90 atomic tasks, 18 short-horizon tasks, 8 long-horizon tasks, and 4 open-ended scenarios.
The authors propose PhysLadder, a hierarchical sim-real benchmark for evaluating physical interaction capabilities in contact-rich manipulation, comprising 16 household tasks divided into three cumulative levels: contact selection and geometric alignment (L1), force regulation under sustained contact (L2), and coordinated contact reconfiguration and force regulation (L3).
The research team presents the BEHAVIOR-Spatial benchmark, built on BEHAVIOR-1K, comprising 3,256 spatial questions across 49 activities, 1,240 grounding queries across 14 manipulation tasks, and action evaluations covering 14 manipulation tasks and 10 navigation tasks.
The authors propose RoboWorM, a deployment-aware benchmark for robotic manipulation video world models that examines structural and spatiotemporal fidelity using hardware priors such as robot kinematic trees and fixed camera configurations. The benchmark collects 1,605 episodes across single-arm, bimanual, and humanoid morphologies, covering Markovian and long-horizon non-Markovian tasks under single-view and multi-view setups, and proposes two deterministic metrics that do not rely on vision-language models: KIVA and RIGIS.
RoboJev-27B is an open-source vision-language model that takes a task instruction and 10 frames of a manipulation attempt as input and predicts whether the robot completed the task. It achieves 95.2% balanced accuracy and 0.993 AUROC on 268 held-out episodes of the same robot model (SO-101), and an AUROC of 0.890 on 459 episodes from unseen labs or robots. Median inference time is 0.26 seconds per episode on a single H100, and the model, inference code, and evaluation results for 20,628 predictions are released together.
The authors propose WorldFailBench, which uses three tasks—failure reason, failure stage, and failure boundary—to test whether world models faithfully simulate policy failures. The benchmark covers manipulation tasks in the LIBERO and RoboTwin simulation environments, as well as real-world manipulation tasks with the G1 humanoid robot. Experiments on seven world models show clear gaps in their understanding of policy failures.
According to Grand View Research's 2025 market study, SoftBank Robotics Group ranks first in the global professional indoor facility mobility robot market (including cleaning robots and indoor delivery service robots) by estimated revenue; it previously held the top spot in 2021.
A CITIC Securities research report concludes that GPT-6 Astra, released by OpenAI on September 3, 2026, is extending from high-level robot task planning toward motion generation and execution control, and that general and specialized models are more likely in the near term to divide work and coordinate.
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 RMMBench, a mobile manipulation benchmark that requires robots to understand language instructions and execute long-horizon tasks in continuous space, with a "navigation-manipulation" task suite covering 70 typical task scenarios. Results show that leading vision-language models (VLMs) still face notable challenges in spatial localization for mobile manipulation tasks, and their spatial perception in long-horizon interactions needs strengthening. The work has been accepted to CoRL 2026.
The study proposes Embodied Agent Arena, evaluating ten VLM agents across five dimensions—geometry, spatial reasoning, affordance, task planning, and manipulation—with 1,000 cases, including the 168-case GeoProbe benchmark for camera, motion, depth, and scale estimation.
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 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.
Ding Ning's team at Tsinghua University, together with researchers from Beijing University of Posts and Telecommunications, Peking University, and Nature Will, built HandsOn, a simulation platform and evaluation benchmark for cross-embodiment dexterous manipulation, containing 66 manually verified tasks and 18,000 human simulation teleoperation demonstrations.
Why it matters: Readers can learn why dexterous manipulation evaluation should distinguish execution from transfer, and which specific failure points current multi-task policies have on bimanual collaboration tasks.
DEEP Robotics has released five research results built around its self-developed embodied brain DEEPRobotics-Brain: the world model DR-WM-1.0, the embodied foundation model DR-VLA-2.0, the embodied agent DR-Agent, the policy optimization framework DR-APO, and the data closed-loop platform DR-Benchmark. Partners include seven universities, among them MIT, Tsinghua University, and Zhejiang University.
DexHoldem, a project involving the University of Hong Kong and Yisheng Technology, brings Texas Hold'em to real hardware, using a UR10e robotic arm and a Shadow dexterous hand to simultaneously evaluate dexterous manipulation, table-state perception, and closed-loop operation. It has been accepted to the NeurIPS 2026 Workshop.
Registration is open for the RoboWorld Challenge 2026, which features three tracks: WorldNav, HA-VLN 2.0, and SafeDrive-VLA. They respectively test language-guided open-loop route planning, social navigation in dynamic crowds, and safe driving under conflicting instructions. The final submission deadline is November 30, 2026.
Limpid Intelligence founder Li Yiming's team at Tsinghua University released the world action model PrismWAM and the execution diagnosis framework RoboDoctor. PrismWAM decouples semantic context from physical state and, without additional action-data pretraining, achieved an average success rate of 83.67% on RoboTwin 2.0 and 75.5% in the Clean-to-Random setting, ranking first on the official leaderboard.
Silicon Valley 101 invited Pokee.ai founder Zhu Zheqing, Pokebot founder Xu Huazhe, and University of Washington assistant professor Wang Sheng to discuss the impact of GPT-6 Astra's release on three paths: personal agents, robotics, and AI drug discovery.