Biorobotics Laboratory BioRob

RB fast learning

Unfortunately this movie is not yet available in a format compatible with recent browsers, and Flash Player is not available or is disabled.. You can try enabling Flash Player in your browser, or as an alternative, you can download the movie (see below) for local viewing with your favorite video player.



This video shows one example of how we automatically generate locomotion patterns for the modular robots by using Roombots modules. Here we have a quadruped robot built by 5 modules. This robot is using oscillatory movements produced by phase oscillator CPG in its 7 degrees of freedom (3 in spine and 4 of the hips). No symmetry is applied for the parameters, therefore the search space has 20 open parameters (7 amplitude, 7 offset and 6 phase differences). The robot is able to find suitable parameters in almost half an hour by just using one pc in the loop (using one processor: Intel(R) core(TM)2 Quad CPU Q9550 @ 2.83GHtz!). The movie shows the initial condition of robot poor-walking, then some iterations during them robot is experiencing different oscillatory patterns in its DoFs sent by optimizer. Robot then sends its speed back to optimizer as the fitness value to improve the generated patterns. At the end, the robot can walk fast by using optimized oscillatory patterns in its Dofs (Work by S. Pouya and J.v.d. Kieboom).