MIT engineers are working on a bipedal robot with highly accurate human reflexes. Named HERMES, the machine can already punch through drywall, swing a baseball bat, smash soda cans, kick over trash buckets, and split a karate board in half.
But here’s the catch. It’s less of an AI (artificial intelligence), and more of a remote controlled robot operated by Joao Ramos, PhD and MIT student. In order to activate and control the robot, PhD Ramos hops onto a special platform and puts on a futuristic-looking exoskeleton made of wires and motors.
Every move PhD Ramos makes is transferred to HERMES in real-time, and the communication goes both ways. When the robot developer mimics throwing a punch, HERMES complies and punches the wall. But once HERMES has punch through the wall, PhD Ramos feels a jolt which triggers his reflex to lean back, making the robot also lean back.
The system is very unique as this is the first robot with a balance-feedback interface. While the robot would still be able to throw punches without this feature, it would not know how to react to the jolt and end up falling straight into the wall.
But the balance-feedback interface allows HERMES to borrow human reflexes from the person wearing the exoskeleton and adjust its balance accordingly. PhD Ramos gave a statement in a video presentation saying that he and his colleagues are “trying to put the human brain inside a robot. So we wanna take advantage of what humans can do, and how humans can learn, and adapt, in order to face new situations and challenges”.
The connection to a human brain also gives HERMES split-second reaction time that a robot would not have otherwise.
PhD Ramos also explained what the problem is with robots that adjust their balance by using the visual feedback provided by onboard cameras. He said that robots typically process images very slowly, so it’s hard for them to react in real-time. But replacing visual feedback from cameras with human reflexes solves that problem.
He went on to give an example: walking. PhD Ramos informed that walking is basically “a process of falling and catching yourself”. And while this is something that humans find effortless, robots developers have found it challenging to program and robots can’t seem to do it both efficiently and dynamically. He said that the next logical step was to let a human take over for the robot.
The main goal of the project is to eventually get HERMES deployed into disasters areas, where it would explore its surroundings while a human operated it from a safe, remote location.
The final stage of the suit would cover the human operator’s entire body, including the eyes (with a pair of goggles), so that he or she can see and feel everything that the robot does.
New robots with the balance-feedback interface would start out as a quadruped before standing up on just two (2) legs in order to perform fairly complex actions such as opening a door or clearing an obstacle.
PhD Ramos and his colleagues are set to make a case for the project next month at the IEEE / RSJ International Conference on Intelligent Robots and Systems.