Tech Xplore Robotics
Covers the latest news on robotics science and technology
HEAPGrasp: A Faster, Smarter Way for Robots to Handle Tricky Objects
Researchers at Tokyo University of Science introduced HEAPGrasp, a novel robot grasping system that relies solely on RGB images to reconstruct 3D shapes of transparent, reflective, and opaque objects. By extracting silhouettes from multiple viewpoints and using a shape‑from‑silhouette algorithm with intelligent viewpoint planning, the method achieves a 96% grasp‑success rate while cutting camera travel by 52% and overall execution time by 19% compared with traditional depth‑sensor approaches. The system can be added to existing robotic platforms without additional hardware, broadening automation possibilities in manufacturing, logistics, and food service. Results were validated across 20 test scenes comprising diverse object types.
AI Benchmark Helps Robots Plan and Complete Their Chores in the Real World
Microsoft and academic researchers introduced GroundedPlanBench, a benchmark that tests robots on 1,009 tasks drawn from 308 real‑world scenarios, exposing gaps between language‑based planning and precise motor execution. To bridge this gap they built V2GP, a video‑to‑spatially grounded planning system...
Digital Twins to Rescue Robots: What Faster 3D Point Cloud Processing Enables
Researchers at Kaunas University of Technology unveiled a hybrid attention‑based model that fuses local detail and global context for 3D point‑cloud segmentation. The system processes complex scenes in just over two seconds per frame while preserving high accuracy, especially for...
Video-Based AI Gives Robots a Visual Imagination
Harvard researchers led by Yilun Du introduced a video‑based robot foundation model that can generate short imagined video clips of future actions. Trained on massive internet video datasets, the system builds a world model that lets robots visualize outcomes before...
AI System Learns to Prevent Warehouse Robot Traffic Jams, Boosting Throughput 25%
MIT researchers and automation firm Symbotic unveiled a hybrid AI system that blends deep reinforcement learning with a fast planning algorithm to prioritize warehouse robots and prevent traffic jams. In simulations mirroring real e‑commerce layouts, the approach lifted robot throughput...
A Robotic Hand without Motors? How a Sub-Second Shape-Shifting Actuator Could Work
A KAIST team led by Professor Seong Su Kim has created a two‑way hybrid actuator that merges shape‑memory alloy (SMA) and shape‑memory polymer (SMP) into a motor‑free device. By embedding a tape‑spring structure, the actuator snaps between bent and flat...
Wristband Enables Wearers to Control a Robotic Hand with Their Own Movements
MIT researchers have unveiled a smartwatch‑sized ultrasound wristband that images wrist muscles and tendons to infer hand gestures in real time. An on‑board AI model translates these images into the 22 degrees of freedom that define finger and palm positions,...
Drones Paired with AI Could Help Search‑and‑rescue Teams Find Missing Persons Faster
Researchers at Kennesaw State University have built an autonomous drone that combines infrared, thermal and color cameras with AI to locate missing persons and assess their condition. The system maps a grid, identifies human bodies, extracts forehead temperature pixel‑by‑pixel, and...
Artificial Neural Network Reproduces Gait Patterns of Four-Legged Animals
Researchers at Brown University's Carney Institute have built a 24‑neuron artificial neural network that can generate five distinct quadruped gaits—bounding, pacing, trotting, walking and pronking—and switch between them instantly. The model leverages attractor‑based dynamics, extending the classic Hopfield framework from...
Five-Level Model Rates Humanoid Robots Across Mobility, Manipulation and Cognition
Fraunhofer IPA introduced the Humanoid Capabilities Navigator, a five‑level model that rates humanoid robots from 0 (absent) to 4 (human‑level) across mobility, manipulation, cognition and safety. The framework lets manufacturers and end‑users compare robots transparently and match capability levels to...
Bird‑like Robots Promise Greater Flexibility and Control than Drones
Rutgers engineers have demonstrated a solid‑state ornithopter that flaps its wings using piezoelectric actuators instead of traditional motors, gears, or linkages. The team also built a multi‑physics computer model that simultaneously captures wing dynamics, aerodynamics, electrical behavior, and control logic,...
Radiation‑hardened Wi‑Fi Chip Survives 500 kGy for Nuclear Plant Decommissioning Robots
Researchers at Institute of Science Tokyo have created a 2.4 GHz Wi‑Fi receiver chip that endures radiation doses up to 500 kGy, a level far beyond typical commercial electronics. By simplifying transistor layouts, enlarging devices, and swapping radiation‑sensitive transistors for passive inductors,...
Insect-Inspired Robot Tracks Odors Even with only One Working 'Antenna'
Researchers have built a bio‑inspired robot that tracks odor sources using a strategy derived from silkworm moths, which can navigate with only one antenna. The platform retains high localization accuracy even when one of its two olfactory sensors fails, performing...
Simple Motor Networks Mimic Human Muscle Behavior Under Increasing Load
Scientists at the University of Bristol have built a tabletop network of simple electric motors that self‑organize into coordinated traveling waves, reproducing the load‑dependent recruitment seen in human muscle. The motors interact only through a shared backbone, creating mechanical feedback...
How Chicago Robot Tutors Are Teaching SEL Effectively, without Pretending to Be Human
University of Chicago researchers partnered with Chicago Public Schools to test robot tutors in fourth‑grade SEL lessons. The experiment compared robots that used fictional, emotion‑laden dialogue with robots that spoke factually and admitted they had no feelings. Both robot groups...
Wind-Powered Robot Could Enable Long-Term Exploration of Hostile Environments
Researchers at Cranfield University unveiled WANDER‑bot, a low‑cost 3D‑printed robot powered entirely by wind. By using a Savonius turbine and Jansen linkage, the robot converts ambient airflow into locomotion, removing the need for batteries. The fully printable design allows on‑site...
Smarter, Faster, and More Human: AI System Helps Robots Outpace Their Human Teachers
Georgia Tech researchers introduced Speed‑Adaptive Imitation Learning (SAIL), a system that lets robots execute learned visuomotor tasks faster than the human demonstrations they were trained on. By dynamically adjusting motion speed while preserving accuracy, SAIL achieved three‑to‑four‑fold speed gains across...
Generative AI Improves a Wireless Vision System that Sees Through Obstructions
MIT researchers have combined generative AI with millimeter‑wave (mmWave) radar to reconstruct hidden 3D objects and entire indoor scenes. The new Wave‑Former system fills gaps in partial reconstructions, boosting shape‑accuracy by roughly 20 % over prior methods. An expanded system called...
Swapping Batteries for Hydrogen Gives Drones a Whole New Range
Researchers at Norway's SINTEF have built a hydrogen‑fuel‑cell drone that swaps its battery for a lightweight hydrogen tank, delivering multi‑hour flight endurance. The prototype can inspect long stretches of power lines, conduct search‑and‑rescue missions, and map terrain, offering a cheaper,...
Sheepdogs Reveal a Better Way to Guide Robot Swarms
Georgia Tech researchers analyzed sheep‑dog trial footage and identified a two‑step herding routine—first subtly aligning stationary sheep, then applying pressure to move them. They built a computational model that captures both the dog’s influence and peer interactions, revealing that small...
Compostable Robot Endures over 1 Million Uses Before Becoming Plant Food
A joint research team from Seoul National University, Sogang University and Johannes Kepler University Linz has created a fully biodegradable soft robot that can endure over one million actuation cycles and then decompose into safe compost. The device uses a...
Autonomous Navigation of Microrobots in Complex Flows Demonstrated for the First Time
Researchers at Leipzig University have, for the first time, demonstrated that synthetic microswimmers can autonomously navigate complex fluid flows by using their own body shape as a sensor. The team employed reinforcement‑learning algorithms coupled with real‑time optical control, enabling particles...
Thermal Cameras Used in Drones and Robots Can Be Tricked by Heat Sources, Study Finds
University of Florida researchers uncovered three previously unknown vulnerabilities in thermal‑camera perception systems used by autonomous drones and robots. The flaws—rooted in image equalization, sensor calibration and lens optics—can be triggered by ordinary or malicious heat sources, causing real obstacles...
4D Printing Technology Uses Waste Sulfur to Enable Self-Actuating Soft Robots
Researchers at KRICT, Hanyang University and Sejong University have unveiled the world’s first 4D‑printing platform that uses sulfur‑rich polymers derived from petroleum‑refining waste. By engineering a loosely cross‑linked sulfur polymer network, the material can be extruded, printed, and programmed to...
New Chip Lets Robots See in 4D by Tracking Distance and Speed Simultaneously
Researchers have created a silicon chip that integrates a 61,952‑pixel focal‑plane array capable of both emitting and receiving FMCW LiDAR signals, delivering 4D imaging that captures distance and speed simultaneously. The prototype generated detailed 3D point clouds at ranges from...
Robots that Learn Everyday Tasks Can Free Humans From Repetitive Work
Korea Institute of Machinery and Materials (KIMM) unveiled the RoGeTA robot task AI, a system that learns everyday activities from human demonstrations and executes them through a hierarchical task‑execution framework. The platform combines task‑extraction, virtualized training environments, and step‑by‑step reasoning,...
A Bicycle Robot that Can Drive Fast and Jump over Obstacles
Researchers at the Robotics and AI Institute unveiled the Ultra Mobility Vehicle (UMV), a bicycle‑based robot that can drive at 8 m/s, hop over 1‑meter obstacles, and perform flips using only five actuated degrees of freedom. The platform uses a heavy...
Hybrid AI Planner Turns Images Into Robot Action Plans
MIT researchers unveiled VLM‑guided formal planning (VLMFP), a hybrid AI system that converts a single image into executable robot action plans. The approach couples a vision‑language simulator (SimVLM) with a generative model (GenVLM) to produce PDDL files for classical planners,...
Poultry Processing Robotics Advances with ChicGrasp
Researchers at the Arkansas Agricultural Experiment Station unveiled ChicGrasp, a dual‑jaw robotic gripper that learns to grasp, lift, and hang chicken carcasses by imitating human motions. The system employs a diffusion‑policy imitation‑learning algorithm and camera perception, achieving an 81% success...
Robot Hands so Sensitive They Can Grab a Potato Chip
Researchers at the University of Texas at Austin unveiled FORTE, a tactile‑sensing robotic hand that can delicately grasp items as fragile as potato chips and raspberries. The soft‑robotic fingers use 3‑D‑printed air channels to detect pressure changes, providing real‑time slip...
New Ultra-Low-Cost Technique Could Slash the Price of Soft Robotics
Oxford engineers have unveiled an ultra‑low‑cost method to fabricate soft‑robotic actuators using vacuum‑sealed plastic pouches and a laser cutter. The process creates functional actuators in under 10 minutes at a material cost of less than $0.10 each. Demonstrations include a...
Researchers Are Combining Drones and AI to Make Removing Land Mines Faster and Safer
Researchers at Rochester Institute of Technology have combined drone‑borne multisensor imaging with artificial intelligence to dramatically speed up land‑mine detection while reducing risk to human deminers. Their field trials showed that drone‑mounted magnetic sensors match ground‑based accuracy and can survey...
Humanoid Robots Master Parkour and Acquire Human-Like Agility
Researchers at Amazon Frontier AI & Robotics and UC Berkeley unveiled Perceptive Humanoid Parkour (PHP), a framework that lets the Unitree G1 humanoid robot perform dynamic parkour using onboard perception. By combining motion‑matching with a teacher‑student reinforcement‑learning pipeline, a single visuomotor...
Bio-Inspired Methods Help Guide Coordination in Underwater Robot Swarms
A new review of 446 papers from 2001‑2025 shows underwater swarm robotics gaining rapid traction. Researchers highlight bio‑inspired coordination rules—drawn from fish schools and predator‑prey dynamics—as a way to overcome the latency and bandwidth limits of acoustic communication. The study...
Will AI Drones, Robots and Wearable Sensors Revolutionize Workplace Safety?
Artificial intelligence‑enabled wearables, drones and robots are poised to transform occupational health and safety, especially in high‑risk sectors such as construction, mining and oil‑gas. Real‑time monitoring of posture, fatigue, noise and hazardous exposure can alert workers before injuries occur, while...
Hardware for Humanoid Robots: New Perspectives for Industrial Value Creation in Europe
Humanoid robots are emerging as a high‑growth market that could outpace the automotive sector, especially in Europe where manufacturers are seeking new value streams. A Fraunhofer‑IPA and P3 white paper reveals a fragmented hardware landscape lacking standardized architectures, with key...
Bioinspired Robot Eye Adjusts Its Pupil to Handle Harsh Lighting
Researchers have built a bioinspired robot eye that uses a liquid‑metal actuator to mimic the pupillary light reflex. The hemispherical sensor array provides an ultra‑wide field of view, while electrical spikes from light‑sensitive material expand or contract the artificial pupil....
Simulated Cats and Elephants with Touch-Based Memory Help Usher in New Age of Robotics
King’s College London researchers unveiled SimTac, a physics‑based simulator that models biomorphic tactile sensors such as cat paws and elephant trunks. By generating virtual training data, SimTac slashes the prototype development cycle from roughly eighteen months to a few weeks....
Brain-Like Chips and LIDAR Sensors May Enable Safer Human-Robot Teamwork
Fraunhofer researchers unveiled NeurOSmart, a system that fuses high‑resolution LIDAR sensing with neuromorphic AI chips to enable safe human‑robot collaboration. The LIDAR sensor uses MEMS mirrors with AlScN material to create a 3‑D view, while AI algorithms pre‑process data directly...
Robotic Wing Inspired by Nature Delivers Leap in Underwater Stability
Researchers at the University of Southampton have created a soft‑robotic wing with a liquid‑metal e‑skin that senses water flow and reshapes itself automatically. In lab tests the wing cut unwanted uplift impulse by 87% versus rigid AUV wings, responded up...
Snake-Like Robot Unveiled for Fukushima Debris Removal
TEPCO unveiled a 22‑metre, 4.6‑tonne snake‑like robot arm designed to retrieve radioactive debris from Fukushima Daiichi. The device can crawl through narrow tunnels, capture visual data, and is slated for a third trial removal later this year. Fukushima still houses...
Soft-Robotic Glove Uses 37 Actuators to Cut Hand Swelling by up to 25%
Cornell researchers unveiled EdemaFlex, a soft‑robotic glove equipped with 37 shape‑memory‑alloy actuators that apply sequential pressure across all five fingers and the palm. In a small home‑use study of seven edema patients, a single 30‑minute session reduced hand volume by...
Researchers Expose Critical Security Vulnerability in Autonomous Drones
UC Irvine computer scientists have uncovered a critical flaw in camera‑based autonomous target‑tracking drones, dubbed the FlyTrap attack. By attaching a specially designed visual pattern to an ordinary umbrella, attackers can trick drones into believing the target is receding, causing...
Rise of the Rice Robots—Creating Active Smart Materials
Researchers at the University of Birmingham discovered that packed rice grains weaken when compressed rapidly—a phenomenon called rate softening. By combining rice‑based granular units with sand, they engineered a speed‑sensitive metamaterial that stiffens or bends depending on loading speed, all...
Engineers Demonstrate Lightweight 'Exoskeleton' That Helps Stroke Survivors Walk
Engineers at the University of Utah unveiled a 5.5‑pound portable hip exoskeleton that reduces the metabolic cost of walking for stroke survivors with hemiparesis by roughly 18%. The device delivers side‑specific motor assistance synchronized in real time, allowing users to...
How Eyes Affect Our Perception of a Humanoid Robot's Mind
Researchers at Tampere University and the University of Bremen used AI‑generated images to compare humanoid robots with and without eyes. Participants consistently rated the eye‑equipped robots as possessing higher agency and experience, regardless of age‑like appearance or eye placement. The...
Humanoid Home Robots Are on the Market—But Do We Really Want Them?
Norwegian‑US startup 1X launched Neo, a $20,000 consumer‑ready humanoid robot that can fold laundry and load dishwashers, but it relies on remote operators wearing VR headsets to handle complex tasks. The robot records video of its actions for future AI...
Humanoid Robots that 'Catch Themselves' Instead of Falling: What a New Walking Algorithm Changes
Georgia Tech researchers unveiled a real‑time planning and control framework that lets bipedal robots recover from instability without human input. Implemented on the Cassie robot, the algorithm uses signal‑temporal‑logic‑guided model predictive control to adjust upcoming steps when perturbations are detected....
How Can You Rescue a 'Kidnapped' Robot? New AI System Helps Robots Regain Their Sense of Location
Researchers unveiled an AI‑driven localization system that lets mobile robots recover their position when GPS signals are lost or degraded. The system fuses data from onboard cameras, LiDAR, and inertial sensors to create a dynamic map and re‑localize the robot...
Robot Hand Approaches Human-Like Dexterity with New Visual-Tactile Training
Researchers in China introduced a visual‑tactile pretraining framework that teaches a low‑cost four‑fingered LEAP Hand to manipulate objects with human‑like dexterity. By first watching extensive videos of human hand actions and then training in simulation using only a webcam and...
