Bipedal robot is hard, how about multiple bipeds working together? Checkout our lab's new research on multi-biped transport. We propose a learned system allowing multiple bipedal robots moving payload together without the need for inter-communication. And sim-to-real!
We explore this question in recent work. Multiple bipeds are arbitrarily attached to a rigid carrier and the carrier velocity is then controlled with a joystick.
We use sim2real RL to train a decentralized controller that is independently run on each biped.
We will also present our work on "Learning Vision-Based Bipedal Locomotion for Challenging Terrain" at ICRA 2024. Check it out at 5/14 at award session on Cognitive Robotics as well as 5/13 at workshop on Agile Robotics: From Perception to Dynamic Action!
We will present our work on "Sim-to-Real Learning for Humanoid Box Loco-Manipulation" at ICRA 2024. Check it out on 5/16 at session Machine Learning for Robot Control II as well as 5/17 at workshop on Loco-Manipulation: Algorithms, Challenges & Applications!
Imagine the possibilities if we can master sim-to-real training of physically capable humanoid robots.
There is a long way to go, but we have our labs first glimmers of success for box loco-manipulation with the Digit V3 humanoid.
https://t.co/EgdJHyhDuV
✨🤖 Robust Standing and Walking: a highly robust humanoid walking controller.
We introduce benchmarks for quantitatively measuring the capabilities of walking controllers.
The results are not trivial, emphasizing why objective measurements are important for our field. ⬇️
In addition to the 5k through campus, our Cassie robot also completed a 5k on an artificial turf surface. We finished with a time just under 44 minutes with no falls! Watch the full run here https://t.co/eLmZ1nuaYG
Congrats to @OregonState Dynamic Robotics Laboratory researchers for achieving a first: making a robot run a 5K! Now that some of these researchers have joined our team, we look forward to incorporating these ideas into commercial applications. #robotics https://t.co/5CNoePFj5k
Our own @JonahSiekmann is going to be presenting "Sim-To-Real Learning of All Common Bipedal Gaits via Periodic Reward Composition" in the Best Paper Award session soon! Tune in to ICRA track 4. His presentation should be from 6:45-7:00 pm PT (9:45-10:00 pm ET)
This account used to be for the ATRIAS robot specifically, but we are now shifting it to be about all the work from the Dynamic Robotics Lab. Expect lots of pictures and videos of our Cassie robot here!