robotic manipulation

Chinese lidar maker Hesai has launched the Kosmo spatial camera, a sensor designed to capture 3D visual data for humanoid robots. The device is intended to give robots better perception of their surroundings, supporting navigation and manipulation tasks. The move marks Hesai's expansion beyond automotive lidar into the fast-growing robotics market, where demand for reliable spatial sensing is rising sharply.

Engineers say one of robotics' toughest problems is not artificial intelligence but the physical act of grasping. A human hand's dexterity, touch sensitivity and ability to adjust grip in real time are extremely difficult to replicate mechanically, making manipulation a bigger hurdle than cognition in many robotics projects.

Robot hands are emerging as a focal point in the robotics industry, with dexterous manipulation seen as a key frontier beyond walking and talking machines. Getting a robotic hand to match human precision and adaptability remains a hard problem, and companies and labs are now competing to crack it first, making the technology a benchmark of progress in the broader race toward capable humanoid robots.

Researchers have unveiled a tiny soft robot that can walk and grip objects without any motors, relying instead on soft, compliant mechanisms for its movement and actuation. The design points to simpler, lighter approaches in soft robotics, with potential uses in delicate manipulation and confined spaces. Details about the team, performance, and applications remain limited in early coverage.

French AI research lab Kyutai has introduced REKI, a 2.4-billion-parameter vision-language-action model aimed at robot manipulation. The model is described as bringing efficient memory use to robotics, allowing robots to perform manipulation tasks with a comparatively small footprint. Coverage so far is limited to a single technical report, so details on performance benchmarks and real-world deployment remain sparse.

Boston Dynamics is highlighting new robotic hand technology designed to pair with modern AI systems and perform practical, real-world tasks. The announcement focuses on giving robots dexterous manipulation beyond walking and mobility, a step toward machines that can handle tools and objects in industrial and everyday settings.

BodySharing Physical AI, described as the world's first physical AI that learns the movements and force control of everyday tasks done with bare hands, including proprioceptive sensing, has been unveiled for the first time at CEATEC in Japan. The announcement is drawing attention as a step toward robots and AI systems that can replicate delicate human manipulation in real-world settings.

Researchers have introduced RoboSimStudio, an interactive toolkit designed for rapid real-to-simulation transfer of robot manipulation tasks. The tool lets robotics developers convert real-world manipulation scenarios into simulated environments quickly, supporting training and testing of robot control policies. Details beyond the toolkit's stated purpose are limited so far, and reactions from the robotics community have not yet emerged.

Researchers have published a new contact-adaptive mesh refinement framework for robotics, described in a paper on arXiv. The approach adds spatial resolution only where it is most useful when manipulating deformable objects, rather than uniformly refining a simulation mesh. The work targets a long-standing challenge in robotic manipulation, where simulating soft, flexible items accurately and efficiently remains difficult.

Boston Dynamics has detailed the new four-fingered robotic hand developed for its Atlas humanoid robot, designed to deliver superhuman grip strength and fine motor control. IEEE Spectrum reports on the engineering behind the hand, which allows the humanoid to manipulate objects with a level of dexterity beyond typical robot grippers, marking another step toward practical humanoids in industrial work.

Researchers have presented a framework for learning team-response beliefs from local history in multi-robot manipulation, published on arXiv in robotics research. The approach addresses how multiple robots coordinating on manipulation tasks can form beliefs about how teammates will respond, based on locally observed interaction history rather than full shared information.

A new arXiv paper in robotics argues that pretrained vision-language models can act directly as manipulation controllers when grounded in an explicit action space, rather than serving only as perception or planning helpers. The approach suggests a path to reusing large pretrained models as the core 'brain' of robotic arms, a step researchers say could cut the cost of training robot-specific policies from scratch.

Researchers have introduced RAIN, Region-Aware Interaction Networks, a robotics approach published on arXiv that frames a robotic task as a sequence of goal-conditioned subtasks, each defined by an action type and a target-region mask. The method aims to give robots more structured, region-focused control over manipulation problems. Discussion is limited so far, with the work circulating mainly among robotics researchers.

Robotics company Agilink has demonstrated a robotic hand learning the fundamentals of embroidery, a task requiring fine dexterity and precise control of fabric and thread. The demonstration highlights progress in robotic manipulation beyond industrial pick-and-place work. Observers see it as a sign that delicate, craft-like manual skills are becoming within reach of automation.

A new survey paper examines how foundation models are being used for affordance learning in robotics, tracing the pipeline from perception through to robotic manipulation. The work reviews how large pre-trained models help robots understand what actions objects afford, consolidating recent progress across vision, language and control. Researchers say the field is moving quickly as general-purpose models are adapted for physical tasks.

Robotics experts are highlighting the gap between humanoid robots' flashy abilities and practical usefulness. While today's machines can sprint, box and perform acrobatic feats that draw viral attention, everyday household tasks like doing laundry remain far harder, requiring fine manipulation, perception and adaptability that current robots still struggle with. Commentators argue that the real measure of progress will be whether robots can handle mundane domestic chores, not just impressive stunts.

Researchers have developed a soft, beetle-inspired robot that can transmit tactile information to its human operator, letting the user sense what the robot is touching in real time. The design combines a flexible, insect-like body with touch feedback, which could make remote manipulation safer and more precise in delicate tasks such as handling fragile objects or exploring hazardous environments.