Self-aware machines are able to assess injuries and make adjustments
Source: “Resilient Machines through Continuous Self-Modeling”
Josh Bongard et al.
Science 314: 1118-1121
Results: By constantly monitoring its own structure, a four-legged robot built by Josh Bongard, a professor of computer science at the University of Vermont, and colleagues at Cornell University can tell if it has damaged or lost a limb and adapt its gait accordingly.
Why it matters: Robots are useful for exploring environments that are too harsh for humans–unless they suffer damage and can’t compensate for it. Previous recovery schemes for damaged robots relied on built-in redundancy such as extra limbs, or preprogrammed contingency plans that anticipated certain failures. Bongard designed a robot that constantly and autonomously monitors itself, adjusting to damage like joint separation or disappearance of a limb. His approach could make robots more useful in harsh environments.
Methods: The robot is equipped with sensors and actuators that collect information about the relative position of its parts. Based on the sensor data, the robot’s onboard computer creates mathematical models of the state of its body. If one of the robot’s limbs is damaged, data from the sensors can be used to generate a new model. A separate algorithm runs simulations of possible gaits, searching for the most efficient one for the damaged robot. The process usually takes about eight hours.
Next steps: Bongard plans to apply his algorithms to a collection of robots. Drawing on the experiences of others in a group could speed up an individual robot’s recovery rate. A damaged robot would send out a query to the other robots in the group, essentially asking if they’d encountered the same injury and how they adjusted.