robotics research

AbuRobot, an Arabic-language robotics reviewer, says he has tested the two best robots in the world in a newly released test video. The footage has drawn close to two million likes, with viewers discussing how today's machines handle balance, movement and real-world tasks. It adds to growing public interest in how far consumer and research robots have advanced.

Researchers have introduced Spa-Bench, a real-robot benchmark for evaluating spatial grounding in vision-language-action models. Results across tested policies show performance drops when robots face out-of-distribution spatial configurations compared to familiar setups, highlighting a key weakness in current systems. The benchmark, described in a new arXiv paper in robotics, gives the field a standardised way to measure how well robot policies understand spatial relationships.

AI company HUMXN has launched a program in Minnesota providing free plumbing and HVAC home service while using the jobs to collect training data for its robots. By having robots work alongside technicians in real homes, the firm aims to teach machines the hands-on skills of skilled trades. The unusual arrangement has drawn attention for blending a public service giveaway with commercial robotics research.

Honda is building a robotic lab assistant designed to support scientific research aboard future commercial space stations. The automaker's robotics division is adapting its technology for microgravity laboratory work, aiming to assist astronauts with experiments as NASA-backed low Earth orbit operations shift from the ISS to privately operated stations.

A new arXiv paper in robotics argues that pretrained vision-language models can act as native action models for robotic manipulation, provided their outputs are grounded in an explicit action space. Rather than training a separate policy from scratch, the approach adapts existing language-vision models to produce robot actions directly. The claim points toward reusing large multimodal models as general-purpose controllers for manipulation tasks.

An education professor is drawing attention for her work bringing robots into teaching, presenting herself and the machines she uses with students. The feature highlights how robotics is being integrated into education and teacher training, and how educators are experimenting with robots as learning tools in classrooms rather than only in research labs.

Honda is developing a robotic laboratory assistant intended to help scientists run experiments aboard future commercial space stations. The robot would support research tasks in microgravity, reducing the workload on astronauts. The announcement has drawn attention as private space stations move closer to operation and companies position themselves for the emerging orbital research economy.

Researchers say first-person, or egocentric, video has become a crucial ingredient for training the AI models that act as the brains for robots. Much as large language models absorb vast amounts of human writing, robot foundation models are increasingly learning from footage captured from a human point of view, helping machines understand how everyday tasks are performed.

Honda is developing a robotic lab assistant intended to operate aboard future commercial space stations, according to Universe Today. The system would support astronauts with laboratory tasks in orbit as low-Earth orbit research shifts from the International Space Station to privately operated facilities. The announcement highlights how automakers and consumer robotics firms are moving into space research support roles.

A new research approach called peering aims to improve how legged exploratory robots interpret their surroundings, letting quadruped 'robot dogs' gather better visual information about areas that are hard to observe directly. The work focuses on visual interpretability, making it easier to understand what the robots perceive as they explore unfamiliar environments.

Researchers publishing in Science have introduced ZEST, a framework that allows robots to transfer learned athletic skills to new tasks without additional task-specific training. The zero-shot embodied skill transfer approach is aimed at improving robot control in dynamic, physical activities, and the paper's appearance in a top journal has drawn attention within the robotics community.

MIT researchers have developed a shape-sensing sheet that digitally tracks its own movement as it bends and twists. The material can sense and record its deformation in real time, a capability that could matter for robotics, wearable devices and soft structures. The news is drawing attention on tech discussion forums, where readers are weighing potential engineering and consumer applications.

DeepSeek has led a $400 million funding round for a robotics AI startup founded by a former Tencent scientist. The deal marks the artificial intelligence firm's first known move into investing, and signals growing capital flowing into embodied AI and robotics ventures founded by veteran Chinese AI researchers. Details about the startup's products and other investors in the round have not yet been fully disclosed.

Q8botOne is being highlighted for progress in open-source quadruped robotics, drawing attention to the growing accessibility of four-legged robot platforms. The development is being covered in technology media as part of broader interest in low-cost, openly available robotics hardware and software. Details about the project's specific capabilities and community response remain limited.

Barclays Investment Bank has published research examining how the rollout of robotics is restructuring economies. The analysis looks at how automation is reshaping industries, labour markets and productivity as machines take on a growing share of work. It has drawn attention from investors and business watchers tracking the economic impact of robotics.

A new article in Nature outlines how biocomputability — the ability of living cells and materials to compute — could be deliberately engineered into biological and synthetic systems. The piece discusses approaches for programming living cells and smart materials to process information, a frontier that could affect medicine, robotics and bioengineering research.

A TechRadar report highlights research into robot-to-robot communication, where a skill or knowledge acquired by one robot can be transferred to others, meaning a single machine's learning could benefit an entire fleet. The idea points toward swarms of robots that update collectively rather than being trained individually, a step with implications for warehouses, factories and autonomous systems.

The Institute for Human and Machine Cognition (IHMC) is developing Alex, a humanoid robot aimed at military applications. The Florida-based research institute is known for advanced robotics work, and the project highlights the growing role of humanoid systems in defense programs. Details about Alex's capabilities and timeline for deployment remain limited.

Engineers have developed a worm-like robot that can travel both on land and in water by shifting fluid inside its own body. By redistributing internal water, the robot changes its shape and weight distribution, allowing it to move across different terrains without external mechanisms. The design draws on how real worms control their body segments, and researchers say it could inspire adaptable machines for environments where wheeled robots struggle, such as coastlines, wetlands and submerged structures.

A new survey paper examines affordance learning with foundation models, tracing how large pretrained models can help robots perceive objects, infer possible actions, and carry out manipulation tasks. It brings together recent research spanning perception and robotic control, offering an overview of how foundation models are reshaping robot learning and what challenges remain before real-world deployment.

A new open-source tool has been released that corrects refraction errors in underwater 3D computer vision. Light bends as it passes from water into camera housings, distorting 3D reconstructions and measurements made underwater. The tool aims to fix these hidden distortions, which matters for marine research, underwater robotics, and subsea inspection work that rely on accurate 3D imaging.

A new open-source framework has been released that simulates drone swarms in fully three-dimensional indoor environments. The tool allows researchers and developers to test swarm coordination and flight behaviour inside detailed virtual indoor worlds before deploying real hardware, reducing cost and risk. Coverage highlights its usefulness for robotics research, indoor navigation testing, and development of autonomous multi-drone systems.

A blog post from security research outlet negativepid walks through how AI systems are constructed, published alongside tags for digital investigations, OSINT, cybersecurity, cyberpsychology and robotics. It is aimed at online investigators who want a technical grounding in how artificial intelligence works and how it may be used or abused in cybercrime contexts.

Taiwan's Industrial Technology Research Institute has showcased a multi-degree-of-freedom serpentine robot as part of its unmanned systems and robotics innovations. The snake-like machine is designed to move flexibly through confined or complex spaces, a capability of interest for industrial inspection, search-and-rescue and other hard-to-reach environments. ITRI positions the prototype among its latest robotics developments.

Research firm IDC reports that Chinese robotics company AgiBot shipped more than 8,600 humanoid robots worldwide in the first half of the year, making it the global market leader by volume. The figure highlights how quickly Chinese firms are scaling embodied AI hardware, outpacing rivals in a sector many see as the next major technology race.

KingKong has released Jumper, an open-source crab robot measuring 40 cm and weighing 1.8 kg. The robot's open-source design means hobbyists and researchers can build, modify and study it themselves, drawing attention from robotics enthusiasts interested in accessible, reproducible hardware projects.

Boston Dynamics has published a new article outlining its work on robotic hands designed to pair modern AI with practical, physical tasks. The piece argues that dexterous manipulation is the next step beyond legged mobility, allowing robots to handle tools and objects in real work environments. It is drawing attention from robotics enthusiasts and researchers debating how far humanoid manipulation has come.