Probabilistically Safe Mobile Manipulation in an Unmodeled Environment with Automated Feedback Tuning

As manufacturing process reconfiguration increases in frequency to support more customized products, manual programming of industrial robots has become less feasible. Although sensor feedback can be used to handle new environments and tasks, continuous sensing is unnecessary for repetitive tasks if a safe trajectory can be learned from prior executions. This paper addresses the problem of developing a controller for a mobile manipulator to safely reach goals without collision while estimating its open-loop safety. To handle novel environments, an iterative feedback tuning scheme is developed based on a fast collision-checking metric and a carefully selected cost function. The open-loop collision probability with the tuned controller is explicitly quantified by propagating uncertainty in the mobile base pose to the sensed environment.