Technical Directions
Robotics coverage relating to manipulation.
390 selected reportsLast 30 days 329 recordsAll coverage 1,104 records
The article reports Geekplus’s proposed RMB 50 million investment for a 20% stake in 深诣机器人 and plans for its embodied-robotics subsidiary to raise outside funding. It also reviews earlier Gino 1 demonstrations and warehouse deployments. The proposed investment is not described as a completed transaction.
Researchers from the University of Tokyo have developed TwinX, a system that uses two robots connected by a shared rope to perform cooperative lifting and transport tasks. The system enables robots to lift a simulated casualty using their legs as arms and then transition into a vehicle-like configuration to transport the casualty. The work demonstrates the potential of multi-robot coordination in complex environments and presents a novel approach to expanding the operational space of cable-driven parallel robots (CDPR).
A team from Tongji University has developed a bio-inspired pulsed arm for quadrotor drones, enabling dynamic aerial interactions while maintaining stability. The design allows for tasks such as grasping, stopping, breaking ice, and cross-media sampling, with the potential to shift control responsibilities from algorithms to mechanical design.
Cainiao Logistics Tech has officially launched its 'vertical leap-type intelligent warehouses' with climbing robots in Tianjin and Suzhou, ahead of the 2026 'Double 11' shopping festival. The system, featuring climbing robots with adaptive adhesion and dynamic balance, significantly improves picking efficiency and reduces manual labor. The deployment is part of a broader trend in China's flexible supply chain, with international expansion planned in Europe and Southeast Asia.
This paper introduces LLM4Robot-Bench, a hierarchical benchmark with 16 household tasks evaluated in matched simulated and real environments. GPT-6 Astra zero-shot achieves 80.0% success at L1 (contact selection/alignment), 47.6% at L2 (force regulation), and 0.0% at L3 (coordinated reconfiguration) in real-world trials, with simulation rates of 66.7%, 47.6%, and 0.0% respectively; ten tasks show identical success counts across environments while six differ by one trial.
This paper introduces PS-DLO, a manipulation policy that combines contrastive image-text learning to infer object properties with SAM3-based mask-guided feature extraction for future observations. Evaluated across nine real-world tasks on single-arm and bimanual platforms, the shared multi-task policy achieves an average success rate of 68.9%, surpassing the strongest baseline by 20.0 percentage points. In generalization experiments, PS-DLO outperforms the baseline in six of seven visual-shift conditions for butterfly-knot untying, demonstrating improved robustness to visual variation when combining object-property priors with future spatial supervision.
This paper presents JAMS, a 3-DoF bionic joint actuator inspired by the human hip, featuring a multi-drive hierarchical output and spherical rotation mechanism. It enables omnidirectional rotation around a fixed center, resolving kinematic redundancy and structural bulkiness. Experimental results show a 360° × 250° × 360° operational space and a 95.1% kinematic-fidelity score in tracking human hip trajectories, demonstrating its feasibility for humanoid and wearable robotics.
Zhuyuan Robotics has released the G2 Max, a safety-compliant heavy-lift embodied robot that combines heavy lifting capabilities with sub-millimeter precision operations. The robot has a single arm payload of 20kg, dual arm rated payload of 40kg, and peak payload capacity of 50kg, with a repeatability accuracy of ≤±0.5mm, making it suitable for various industrial applications to enhance production line efficiency and space utilization.
Zhiyuan officially launched the G2 Max, an industrial-grade heavy-duty embodied robot with 20kg single-arm load capacity, 40kg dual-arm load capacity, and 50kg peak load capability. It achieves a repeat positioning accuracy of ±0.5mm. The robot has already been tested and applied on three real production lines, covering high-end manufacturing, warehousing logistics, and fast-moving consumer goods scenarios. It unifies heavy lifting with precise operations, improves factory space utilization, and supports 7×24 continuous operation.
CASBOT says it completed an A++ financing round of several hundred million yuan. It reports shoe-manufacturing orders and mine trials, with proceeds allocated to model development, delivery and market expansion. Financial and performance figures are company-reported.
The 'RoboMIND' dataset, developed by Beijing Humanoid Robotics Innovation Center, has surpassed 50 million global downloads since its open-source release in December 2025. It is the first dataset to support dual-arm coordination, mobile manipulation, dexterous hands, and high-fidelity tactile perception, enabling comprehensive training for robotic tasks. The dataset includes over 400,000 dual-arm operation trajectories and covers more than 700 real-world tasks.
The article reports on the mass production of personal robots, with the 5000th unit being manufactured in under a month, achieving a production rate of one robot every 2.5 minutes. It also highlights the world record for 110 humanoid robots performing synchronized dance, showcasing the scalability and coordination of production-ready robots. Additionally, it discusses the deployment of robots in industrial settings, emphasizing real-world applications and the transition from lab research to practical use.
This research introduces WarpWeaver, a decentralized, real-time planner that approaches joint optimality for multi-robot nonlinear systems by using dispersive trees of forward-propagated commands and negotiation driven by decentralized column generation. The method replaces inter-robot coupling with local constraints, allowing locally grown dispersive trees to contain near-optimal agreements. In simulation on the Dynobench suite, WarpWeaver reduces total travel time by up to 22% compared to centralized offline planners and by up to 16% against a decentralized real-time baseline, while outperforming task-specific planners on networked navigation. Hardware experiments on six robots planning onboard demonstrate real-world implementability.
This research proposes a dual-loop assistance strategy for exoskeletons that couples multimodal trajectory generation with variable admittance control. The method uses whole-body human motion and exos, and incorporates task descriptions as semantic cues. A three-stage training pipeline reduces cross-modal representation mismatch, while trajectory variance provides an uncertainty measure for prediction reliability. Experimental results show reduced muscle activation and metabolic energy expenditure compared to standard baselines.
This paper presents WholeBodyMem, an object-guided perceptual memory system that retains pre-history information for humanoid loco-manipulation in dynamic environments. By using SAM3 to track task objects and causally selecting key frames within an adaptive budget, the action expert integrates these frames via cross-attention with object tokens encoding appearance, geometry, and time. The authors also release HuMeGo, a benchmark featuring six simulated and six real-world Unitree G1 tasks. WholeBodyMem improves success rates from 29.1% to 72.6% in simulation and from 36.7% to 88.3% on the real robot, while also enhancing performance on standard SIMPLE tasks from 70.2% to 86.9%.
A preclinical animal study conducted by Shenzhen Institute of Advanced Biomedical Robot Co., Ltd. in collaboration with Professor Jianmin Liu's team at Shanghai Changhai Hospital has validated the capabilities of the neurointerventional robotic system PANVIS STAR®. The system successfully completed five major procedures, including cerebral angiography, aneurysm coil embolization, flow diverter deployment, vertebral artery stenting, and carotid artery stenting, with no device-related complications or vascular injury. The study, published in the Journal of Robotic Surgery, marks a significant step toward treatment-capable robotic systems in neurointerventional procedures.
Robotiq has launched the 2FC physical AI gripper, a production-ready adaptive gripper with tactile sensing and embedded intelligence. It captures and labels contact-rich data for training and runs embedded intelligence locally to enable real-time robotic action based on tactile feedback. The gripper is available for pre-order in two stroke versions, with shipping beginning in Q1 2027.
This research paper introduces Agi, a large-scale dataset with over 1 million trajectories across 217 tasks, and presents Genie Operator-1, a policy that achieves 30% performance improvement over existing methods. The dataset and tools are open-sourced to advance research in scalable embodied intelligence.
Mitsubishi Electric has developed the industry's first machining system that uses physical AI to coordinate humanoid robots with CNC machine tools, enabling humanoids to perform tasks such as setup, cleaning, and inspection. The system uses natural language for communication between operators, machines, and robots, reducing the need for custom integration and supporting unmanned machining operations.
The article explores the deployment of logistics robots in warehouse sorting tasks, focusing on non-humanoid designs that are optimized for high-frequency, data-rich environments. It discusses the challenges of human labor in sorting, the technical and economic rationale for robotic solutions, and the evolving role of embodied AI in logistics automation.
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).