Vision-based motion planning of a pneumatic robot using a topology representing neural network

We present a new approach to integrate sensors into robot motion planning by combining the concept of the perceptual control manifold (PCM) and the topology representing network (TRN) algorithm. Motion planning should incorporate sensing due to the presence of uncertainty. Therefore, the PCM extends the notion of robot configuration space to include sensor space. Exploiting the topology preserving features of the TRN algorithm, the neural network learns a representation of the PCM. The learnt representation of the manifold is then used as a basis for motion planning with various constraints. The feasibility of this approach is demonstrated by experiments with a pneumatically driven robot arm (SoftArm)