Video Friday is your weekly collection of superior robotics movies, collected by your folks at IEEE Spectrum robotics. We additionally put up a weekly calendar of upcoming robotics occasions for the subsequent few months. Please send us your events for inclusion.
ICRA 2026: 1–5 June 2026, VIENNA
Get pleasure from right this moment’s movies!
Not like present hybrid designs, Duawlfin eliminates the necessity for added actuators or propeller-driven floor propulsion by leveraging solely its customary quadrotor motors and introducing a differential drivetrain with one-way bearings. The seamless transitions between aerial and floor modes additional underscore the practicality and effectiveness of our method for functions like city logistics and indoor navigation.
[ HiPeR Lab ]
I recognize the softness of NEO’s design, however these fingers look awfully fragile.
[ 1X ]
Think about reaching into your backpack to search out your keys. Your eyes information your hand to the opening, however as soon as inside, you rely nearly completely on contact to differentiate your keys out of your pockets, telephone, and different gadgets. This seamless transition between sensory modalities (understanding when to depend on imaginative and prescient versus contact) is one thing people do effortlessly however robots battle with. The problem isn’t nearly having a number of sensors. Fashionable robots are geared up with cameras, tactile sensors, depth sensors, and extra. The actual drawback is **tips on how to combine these totally different sensory streams**, particularly when some sensors present sparse however important info at key moments. Our resolution comes from rethinking how we mix modalities. As an alternative of forcing all sensors via a single community, we prepare separate knowledgeable insurance policies for every modality and learn to mix their motion predictions on the coverage stage.
Multi-university Collaboration introduced through [ GitHub ]
Thanks, Haonan!
Completely satisfied (considerably late) Halloween from Pollen Robotics!
[ Pollen Robotics ]
In collaboration with our colleagues from Iowa State and College of Georgia, we’ve got put our pipe-crawling worm robot to check within the subject. See it crawls via corrugated drainage pipes in a stream, and a clean part of a subsurface drainage system.
[ Paper ] from [ Smart Microsystems Laboratory, Michigan State University ]
Heterogeneous robotic groups working in sensible settings typically should accomplish complicated missions requiring collaboration and adaptation to info acquired on-line. As a result of robotic groups incessantly function in unstructured environments — unsure, open-world settings with out prior maps — subtasks have to be grounded in robotic capabilities and the bodily world. We current SPINE-HT, a framework that addresses these limitations by grounding the reasoning talents of LLMs within the context of a heterogeneous robot team via a three-stage course of. In real-world experiments with a Clearpath Jackal, a Clearpath Husky, a Boston Dynamics Spot, and a high-altitude UAV, our technique achieves an 87% success charge in missions requiring reasoning about robotic capabilities and refining subtasks with on-line suggestions.
[ SPINE-HT ] from [ GRASP Lab, University of Pennsylvania ]
[ Astribot ]
In two papers revealed in Matter and Superior Science, a workforce of scientists from the Bodily Intelligence Division on the Max Planck Institute for Intelligent Systems in Stuttgart, Germany, developed management methods for influencing the movement of self-propelling oil droplets. These oil droplets mimic single-celled microorganisms and may autonomously resolve a posh maze by following chemical gradients. Nonetheless, it is vitally difficult to combine exterior perturbation and use these droplets in robotics. To deal with these challenges, the workforce developed magnetic droplets that also possess life-like properties and may be managed by exterior magnetic fields. Of their work, the researchers confirmed that they’re able to information the droplet’s movement and use them in microrobotic functions akin to cargo transportation.
Everybody has fantasized about having an embodied avatar! Full-body teleoperation and full-body data acquisition platform is ready so that you can strive it out!
[ Unitree ]
It’s not a humanoid, however it proper now safely does helpful issues and possibly doesn’t price all that a lot to purchase or run.
[ Naver Labs ]
This paper presents a curriculum-based reinforcement learning framework for coaching exact and high-performance leaping insurance policies for the robotic `Olympus’. Separate insurance policies are developed for vertical and horizontal jumps, leveraging a easy but efficient technique. Experimental validation demonstrates horizontal jumps as much as 1.25 m with centimeter accuracy and vertical jumps as much as 1.0 m. Moreover, we present that with solely minor modifications, the proposed technique can be utilized to study omnidirectional leaping.
[ Paper ] from [ Autonomous Robots Lab, Norwegian University of Science and Technology ]
Heavy payloads aren’t any drawback for it: The brand new KR TITAN extremely strikes payloads of as much as 1500 kg, making the heavy lifting excessive within the KUKA portfolio.
[ Kuka ]
Good luck getting all the sand out of that robotic. Maybe a pleasant oil tub is so as?
[ DEEP Robotics ]
This CMU RI Seminar is from Yuke Zhu at College of Texas at Austin, on “Towards Generalist Humanoid Robots: Latest Advances, Alternatives, and Challenges.”
In an period of fast AI progress, leveraging accelerated computing and big data has unlocked new prospects to develop generalist AI models. As AI techniques like ChatGPT showcase outstanding efficiency within the digital realm, we’re compelled to ask: Can we obtain related breakthroughs within the bodily world — to create generalist humanoid robots able to performing on a regular basis duties? On this discuss, I’ll define our data-centric analysis ideas and approaches for constructing general-purpose robotic autonomy within the open world. I’ll current our latest work leveraging real-world, artificial, and net information to coach foundation models for humanoid robots. Moreover, I’ll focus on the alternatives and challenges of constructing the subsequent technology of clever robots.
[ Carnegie Mellon University Robotics Institute ]
From Your Website Articles
Associated Articles Across the Net
