Researchers at the
Ray Institute are exploring one of the most complex challenges of modern robotics, teaching machines to manipulate objects with their entire bodies in a fluid and coordinated way, not only with their hands, but also with their arms, legs and trunk, as humans and some animals do.
The goal is to overcome the limitations of traditional robotics in which robots grab and move objects statically and achieve truly dynamic behavior where robots can push, roll, hit, lift and balance objects with their own mass, inertia and kinetic energy. To achieve this, scientists created a hierarchical control architecture.
At the base, a deep reinforcement learning model guarantees balance and stability, controlling motor torques in real time, above it, the high-level control is responsible for understanding the task such as stacking, dragging or rolling a tire and planning handling strategies, so it was able to discover for itself the most effective strategies, how to use the legs and trunk as levers to lift a 15 kg tire, something well above its nominal load capacity.
The system was tested in tasks such as rolling tires, stacking and dragging, with an interesting performance, the robot completed the lifting of a tire and an average of 5.9 seconds, reaching in some executions the average reference time of a human. The big difference in this research lies in the integration between locomotion and manipulation, instead of treating, walking and manipulating as separate actions, the unified control allows the robot to combine both naturally, something essential for real tasks such as moving furniture, carrying boxes or assembling structures.
This fusion breaks the paradigm of almost static manipulation, predominant in robotics for decades, where movement is slow and limited. The new system is active and physical, using acceleration, inertia and multiple contact with the environment to achieve human performance. Despite advances, the system still relies on external motion capture and computing cameras outside the robot; next stages include replacing these dependencies with built-in vision, touch sensors, and fully autonomous control.
Thanks for your contribution to the STEMsocial community. Feel free to join us on discord to get to know the rest of us!
Please consider delegating to the @stemsocial account (85% of the curation rewards are returned).
Consider setting @stemsocial as a beneficiary of this post's rewards if you would like to support the community and contribute to its mission of promoting science and education on Hive.