IEEE Spectrum’s weekly ‘Video Friday’ column gathers a selection of robotics videos, including a demonstration of ‘whole-body’ manipulation conducted by the Robotics and AI Institute and a humanoid mobility research system. The issue also mentions a new vision board, modular robotic hands, and a test robot designed for industrial pipelines.
Humanoid manipulation and mobility
The Robotics and AI Institute (RAI) presents a quadruped robot manipulating a tire on an outdoor basketball court, using its entire body. In its blog post, the institute describes a challenge: ‘Anything that appears as a blob in a depth image can be mistaken for the tire, whether it’s a person passing by or an object left on the ground. In a few trials, the robot dropped the tire to go after something else.’ The video does not allow for certainty on whether the torso participates in the grasp.
Meanwhile, the Robotic Systems Lab at ETH Zurich presents a contact-centric learning framework for versatile mobility of humanoids in complex, unstructured environments. According to the team, the approach combines motion imitation and reinforcement learning, i.e., a method where the system learns through trial and error, using human motion data as a starting point. It would also allow for autonomous adaptation to previously unseen terrains.
Sensors, hands, and platforms
Luxonis launches the OAK 4 FFC, a 70 x 53 mm vision computer that allows placing up to five cameras where the robot needs them. According to the company, the board handles synchronization, stereo depth, and artificial intelligence itself, and calibrates from the scene rather than using a checkerboard pattern.
Robotis presents HX Friends, a series of robotic hands powered by DYNAMIXEL, ranging from a single articulated finger to a gripper, then to three-, four-, and five-fingered hands. The video shows movement demonstrations and teleoperation in virtual reality of the multi-fingered hands, i.e., their remote control by an operator. Robotiq, for its part, unveiled a new tactile-sensing gripper.
Developer bluZenit presents monkeyBot 2.0, a platform designed to move through industrial piping networks. The robot uses a gripping and climbing mechanism to travel on vertical and horizontal pipes, navigate elbows, and access hard-to-reach areas.
Research and competitions
The MAVLab team from TU Delft participated in April 2025 in the A2RL drone racing championship in Abu Dhabi. Its AI algorithm, MonoRace, won the AI Grand Challenge, then defeated three human FPV piloting world champions. The drone uses only a single onboard camera and an inertial measurement unit, a motion sensor, to estimate its state. An embedded neural network directly converts this estimated state into motor commands.
Another publication presents SAWbot, a bio-inspired amphibious robot that swims using a continuous wave-like motion, powered by a single motor. According to the paper’s authors, it is among the most energy-efficient swimming robots published to date. It can also walk on land and transition from water to land with minimal disturbance to aquatic life.
Finally, the column reminds readers of upcoming events: the CoRL 2026 conference, from November 9 to 12, 2026 in Austin, and Humanoids 2026, from December 6 to 9, 2026 in Santa Clara, California.
