Last edited:
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
RoboCup 2017 in Nagoya Japan was held
with robots and researchers all around the world.
ROBOTIS look into the humanoid soccer games
and behind the scenes.
Robotics developer Boston Dynamics debuted it's humanoid Atlas in a demo with another robot, Spot. At the end of the demo, Atlas fell walking off the stage to produce an epic fail!
Boston Dynamics Atlas robot failing and falling doing parkour.
It's hilarious to see a robot trip and fall. Boston Dynamics has released a video of its multimillion-dollar Atlas robot failing and falling at parkour, which, it turns out, happens significantly more frequently than a good run.
Designing a robot that is as nimble as a human is a tremendous job, and programming it to behave and move like a human is much more difficult. Atlas is one of the most complicated robots ever built, and as Boston Dynamics detailed today, anything that can go wrong will, including random hydraulic fluid leaks and bolts coming off as the 190-pound robot makes harsh landings.
Atlas' inventors, on the other hand, are prepared for such blunders; it's all part of being on the bleeding edge of robotics and testing a robot that's primarily intended as a research tool. It breaks, and it breaks frequently, but each time it does, it provides an opportunity for Atlas to enhance its design so that, in another year, the business will be able to produce videos displaying the bot's incredible new abilities. Robot crashes are all part of the learning process for Boston Dynamics engineers and technicians. Breaks, bumps, and failures serve as reminders of what has to be changed, allowing the team to design a better robot.
We propose a robot agnostic reward function that balances the achievement of a desired end pose with impact minimization and the protection of critical robot parts during reinforcement learning. To make the policy robust to a broad range of initial falling conditions and to enable the specification of an arbitrary and unseen end pose at inference time, we introduce a simulation-based sampling strategy of initial and end poses. Through simulated and real-world experiments, our work demonstrates that even bipedal robots can perform controlled, soft falls. Publication link: https://arxiv.org/abs/2511.10635