Reservoir Computing Enables Machine Learning to Turn a PAM Into a Combined Muscle and Sensor

Adding carbon to common pneumatic artificial muscles (PAM) turns them into a source of data for a machine learning system to model.

A "reservoir computing" system turns a common, and cheap, soft robot muscle into a sensor for modeling. (πŸ“·: Sakurai et al)

Researchers at the University of Tokyo have unveiled a machine learning system capable of using soft materials as both a muscle and a sensor for flexible robots, building on earlier work inspired by octopus tentacles.

Soft robotics is a rapidly-advancing field, but getting both sensing and actuation in a flexible form is a challenge β€” and one researchers working in the University of Tokyo's Next Generation Artificial Intelligence Research Centre have addressed, looking to resolve the issues with the difficult-to-predict nature of soft robotic systems.

β€œTake for example a robot with pneumatic artificial muscles (PAM), rubber and fibre-based fluid-driven systems which expand and contract to move," explains Associate Professor Kohei Nakajima from the Graduate School of Information Science and Technology. "PAMs inherently suffer random mechanical noise and hysteresis, which is essentially material stress over time. Accurate laser-based monitors help maintain control through feedback, but these rigid sensors restrict a robot's movement, so we came up with something new.

"We found the electrical resistance of PAM material changes depending on its shape during a contraction. So we pass this data to the network so it can accurately report on the state of the PAM. Ordinary rubber is an insulator, so we incorporated carbon into our material to more easily read its varying resistance. We found the system emulated the existing laser-displacement sensor with equally high accuracy in a range of test conditions."

The result is an artificial muscle which doubles as a sensor, with the neural network using the concept of reservoir computing to constantly take in about the PAM's condition and change its conceptual model on-the-fly. "Our study suggests reservoir computing could be used in applications besides robotics," Remote-sensing applications, which need real-time information processed in a decentralized manner, could greatly benefit," continues Nakajima. "And other researchers who study neuromorphic computing β€” intelligent computer systems β€” might also be able to incorporate our ideas into their own work to improve the performance of their systems."

The team's work has been published in the journal Proceedings of the IEEE International Conference on Soft Robotics (RoboSoft) 2020, but is not yet publicly available; more information is available on the University of Tokyo website.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire:
Related articles
Sponsored articles
Related articles