HERMES is a two-legged robot that MIT researchers have enabled to carry out human movements, through a unique balance-feedback interface.

HERMES can punch through drywall, smash soda cans, kick trash bins and even karate-chop boards in two. These actions aren't, however, directly executed by HERMES. Instead, a researcher strapped into an exoskeleton with motors and wires is controlling the robot by performing the very same movements. As the researcher simulates punching through a wall, the robot will do the same. But as HERMES' fist hits the wall, the researcher will also feel the impact, and push back to maintain balance — which will be copied by HERMES.

The balance-feedback interface utilizes the split-second reflexes humans have, giving HERMES a faster reaction time than if it were to adjust its footing according to visual information from installed cameras. Joao Ramos, a PhD student from the MIT Department of Mechanical Engineering, said that image processing is slow and this makes it difficult for a robot to react in time. He added that the researchers are keen on exploring how people can take over movements for a robot.

With the balance-feedback interface, Ramos and his colleagues are hoping to deploy HERMES to disaster sites, where it can explore the affected area with the guidance of a human operator.

"We plan to have the robot walk as a quadruped, then stand up on two feet to do difficult manipulation tasks such as open a door or clear an obstacle," Ramos explained.

The balance-feedback interface acts as a mediator between HERMES and its human controller. To sense the robot's balance, the researchers fitted HERMES with load sensors capable of measuring force exerted by each of its feet as it takes a step. Calculating where the robot has shifted its weight clues the controller in if the robot is about to fall, allowing for corrective measures in order to regain balance.

The next goal is to have HERMES do more complex movements, like opening a spring-loaded door or swinging an axe without losing its footing.

The study received funding support from the Defense Advanced Research Projects Agency. It will be presented by Ramos at the IEEE/RSJ International Conference on Intelligent Robots and Systems along with colleagues Sangbae Kim and Albert Wang.