Papers & Evaluations Latest coverage

ForeAct3D: Policy-Grounded Future World Modeling for VLA Policies

This paper presents ForeAct3, a framework for policy-grounded future world modeling within VLA policies. It uses learnable geometric queries to decode depth, semantic segmentation, and camera pose from policy representations, enabling accurate prediction of future 3D scene states. The method improves manipulation performance through semantic 3D supervision, physical consistency, and action conditioning, achieving 98.3% average success on LIBERO and 37.8% success in real-world tasks.

NovaSplat: Model-Free 6D Object Pose Estimation via 3D Gaussian Splatting Onboarding and Refinement

NovaSplat is a model-free pipeline for 6D object pose estimation that trains a 3D Gaussian Splatting model from a short smartphone video. It serves as both a template for detection and a differentiable backbone for pose refinement, outperforming model-based methods on a household-object benchmark and model-free competitors on an industrial benchmark with textureless and specular surfaces. A faster inference variant, NovaSplat-Lite, is also introduced.

When a Robot Reaches for a Cup, It's Moved Away: LIBERO-MAX Uses 224,000 Simulations to Test Dynamic Reliability

LIBERO-MAX is a study that evaluates the adaptability of robots in dynamic environments by designing paired tasks and mid-task changes to assess the performance of different strategies under environmental changes. The study found that the success rates of all tested strategies decreased, and the impact of different types of changes on the models varied significantly, indicating that robots need more robust perception and planning mechanisms to handle uncertainties in real-world scenarios.

USTC Team Puts a 'Reverse Thinking' World Model into Robots, Secures Millions in Funding in Just Three Months

Bai Ze Tong Jing is a less than six-month-old company based in Hefei, and its developed Causal World Model AWM enables robots to perform causal reasoning about environmental changes. AWM significantly improves data efficiency and task success rates by learning object, motion, and interaction patterns from unlabeled videos.

Are robots better at soldering? A study verifying whether 'learning' in industrial robots can change factories

A research paper explores the use of learning-augmented robotic automation in industrial settings, focusing on a three-phase cable insertion and soldering process. The hybrid system combines routine movements with learning-based control, using visual servoing and safety monitoring to achieve a 99.4% success rate in a factory test. While the robot is slower than humans, its consistency and safety features suggest potential for future industrial applications.

Weiyi Zhizao Passes HKEx Listing Hearing: Calculating the ROI of Industrial Embodied Intelligence

Weiyi Zhizao has passed the Hong Kong Stock Exchange listing hearing, marking a shift in the industrial embodied intelligence sector from conceptual validation to real-world ROI. The company's financial performance, including revenue growth, improved margins, and customer adoption, demonstrates its ability to deliver scalable solutions. Its 'Physical AI Brain' technology, combining AI models, industrial expertise, and real-world data, enables continuous improvement and deployment efficiency, forming a positive feedback loop between technical capability and commercial success.

Add a Stool if You Can't Reach. Li Hengyang's Team Launches Real-World Robot Achievements After Six Months of Startup

The Yuance Future team demonstrated their robot's performance in complex tasks, including reaching high places to retrieve objects, picking up bottles from low positions, clearing tables, and operating dishwashers. The article introduced their 'Whole-Body Intelligence' (Whole-Body Intelligence, WBI) technology approach, emphasizing the integration of body coordination and task execution. By collecting human motion data, real-machine training, and cross-body validation, the robot gradually adapts to various challenges in real-world scenarios.

Hangzhou's Embodied Robot Rivals Figure, Demonstrates Full-Body Autonomous Clothing Folding

The article describes how the WR1 robot, developed by West Lake Robotics, successfully performs complex household tasks such as folding clothes and managing multi-scene operations. It highlights the use of a dual pre-training system, world models, and a general-purpose action execution model to achieve seamless coordination between perception, planning, and execution. The robot's ability to handle flexible objects and maintain task continuity across different environments is emphasized as a significant breakthrough in embodied AI.

Robotic Systems Lab@leggedrobotics· 04:32SignalEditorial score6868

Researchers from the ETH Robotic Systems Lab present a world model for quadruped robots that allows zero-shot policy transfer across different robot morphologies without fine-tuning, retraining, or warm-up. The model conditions policies on dynamics rather than directly training policies, enabling generalization across heterogeneous quadrupeds.

Two-armed robot throws and catches balls with human-like movements

AthenaZero, a new dual-arm robotic system developed by researchers at the Robotics and AI Institute, demonstrates agile, human-like movements in tasks such as throwing, catching, and batting. The robot's design incorporates low inertia and backdrivability, enabling more natural and controlled interactions with objects. The study explores the challenges in robotic manipulation and presents a novel approach to improving dexterous interaction with the environment.

LeRobot@LeRobotHF· 13:34SignalEditorial score7272

LeRobot's new streaming loader is designed for training on large-scale robot datasets, allowing efficient processing of 100 TB+ data without full local video copies. It reads episodes directly from Hugging Face or Buckets, prepares samples on-the-fly, and balances memory and throughput for varied training batches.

Humanoid Robot Uses Dexterous Hand to Fold Clothes Autonomously, West Lake Robot Achieves It

West Lake Robot's WR1 humanoid robot demonstrates the ability to autonomously fold clothes using its dexterous hand, marking the first global instance of such capability. The robot performs a series of complex tasks, including retrieving corn from a fridge, moving to an air fryer, organizing toys, and folding clothes, all while maintaining smooth motion and avoiding task interruptions. This is achieved through a novel 'general brain' architecture that integrates perception, decision-making, and motion control, enabling seamless, dynamic task execution in real-world environments.

Tsinghua's Embodied Model Tops Global Rankings. Outperforming GPT-6 and NVIDIA Without Hacks or Extra Data

Xingdong Jiyuan's self-developed world action model VPP2 has topped the RoboDojo simulation leaderboard with an average success rate of 32.26% and a score of 39.26, outperforming models like GPT-6-Astra. VPP2 improves prediction and action generalization by decoupling video prediction from action learning through a phased training strategy.

Aether AI's CRIS-0: Causal-Driven Robot Intelligence for Real-World Adaptation

Aether AI introduces CRIS-0, a causal-driven robot intelligence system that enables robots to adapt to unexpected disturbances in real-world environments. The system uses causal reasoning to understand task-relevant variables, predict the consequences of actions, and adjust strategies dynamically. CRIS-0 is structured into two layers: a high-level agent for task planning and a causal world model for predicting physical consequences. The research demonstrates the effectiveness of causal reasoning in dynamic, human-robot collaborative settings, with experiments showing improved robustness and task completion compared to traditional prediction-based approaches.

LM-X: Explainable Vision--Language--Action Modeling via Progress, Event, and Uncertainty Prediction

LM-X is a new VLA foundation model that learns three directly supervised online signals—return-to-go (RTG), event-to-go (ETG), and heteroscedastic action-flow variance—to provide explainable control. Pretrained on over 20,000 hours of real-robot trajectories, LM-X achieves 74.1% success on RoboTwin2.0 tasks and outperforms GR00T N1.7 in both simulated and real-world settings. The model's signals track progress, anticipate motion, and detect high-error actions, establishing it as a transparent and effective VLA framework.

ARC: A Reasoning Recipe for Robot Foundation Models

This research proposes ARC, a reasoning framework that improves zero-shot task performance of existing robot foundation models (RFMs) through structured reasoning traces, automatic labeling, and model adaptation. The method achieves state-of-the-art results on benchmark tasks like RoboLab-120 and MolmoSpaces, with up to 50 percentage point gains on RoboLab-Reasoning-50 and an 82.2 percentage point improvement in task success on real robots.

NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime

NavGPT-3 connects language models with action policies through a hierarchical runtime system, enabling autonomous agents to navigate complex environments. The system uses a reasoning loop and an action policy trained on 19.28M examples, achieving state-of-the-art results on R2R-CE and matching human performance on RxR-CE. The paper also explores the interaction between the models and the impact of tools on navigation performance.

Multi-Agent Coordination via Support-Preserving Distillation

This research paper addresses failure modes in offline multi-agent reinforcement learning (MARL) where standard flow-based teachers route noise samples to conflicting coordination modes. The authors propose Mode-Support Semi-Discrete Optimal Transport (MoSDOT), a method that summarizes multimodal replay into finite mode support with prescribed capacities and uses conditional semi-discrete optimal transport to assign each noise sample to a single mode before teacher training. Additionally, a shared-randomness variant is studied using a shared noise component at execution to expose the residual gap intrinsic to strict-product execution. On controlled diagnostics and offline MARL benchmarks, MoSDOT improves endpoint quality and routing consistency, particularly on datasets exhibiting multimodal joint behavior.

Long-WAM: Scaling the Context of World-Action Models

This research presents Long-WAM, a model-system framework that scales the context of causal world-action models under real-time control constraints. By using autoregressive pretraining, the model improves task success rates on benchmarks like RoboCasa GR-1 and LIBERO-Long. It also supports real-time deployment on hardware platforms, enabling dynamic and long-horizon manipulation tasks.

MagCilia: A Compact Magnetociliary Tactile Sensor with 3D Force Sensing for Robotic Contact Perception and Grasping Feedback

This paper introduces MagCilia, a compact tactile sensor that integrates a flexible magnetic-cilia structure with a Hall sensor to enable three-dimensional force sensing. Using quasi-static finite element analysis and causal history fusion regression (CHFR), the system reconstructs forces from coupled magnetic channels with root-mean-square errors of 0.40 N, 0.57 N, and 0.69 N for Fx, Fy, and Fz respectively, achieving coefficients of determination between 0.90 and 0.93. Robotic experiments demonstrate the sensor's ability to guide gripper adjustments based on tangential forces and monitor multi-axis loads under external perturbations. Additionally, frequency-domain features allow the system to distinguish six surface categories with 99.39% accuracy in cross-validation grouped by acquisition session, and an online demonstration successfully identifies all six tested surfaces.