Programming Mobile Robots by Demonstration through System Identification

For certain mass-marked robot applications it is essential to have cheap, efficient and reliable methods of sensormotor code generation, hand-programming is not a viable option in such cases. We present a new method to program mobile robots — “code identification by demonstration” — which algorithmically transfers human behaviours directly to robot control code, using transparent mathematical functions. In contrast to traditional robot programming techniques, our method does not require specialised technical or programming skills, but translates demonstrated behaviour immediately into executable code. This has obvious implications for the widespread use of personal service robots. The proposed method is fast and computationally cheap, and has the additional advantage of generating transparent, i.e. analysable controllers using polynomials, which permits hypothesis formation and theoretical analysis of robot behaviour, for example for stability and safety analyses. We demonstrate the viability of this approach by teaching a Scitos G5 mobile robot to achieve corridor following behaviour.