Task Planning With Active Sensing For Autonomous Compliant Motion

Previous research showed that the execution of contact tasks under uncertainty benefits from online estimation of the geometrical contact parameters such as positions, orientations and dimensions of the contacting objects. However, the constant translational and rotational velocities commonly used to trigger the contact formation (CF) transitions are often not sufficiently exciting to estimate all geometrical parameters. In this paper, we focus on the calculation of a fine-motion task plan which improves the observation of inaccurately known geometrical parameters. This is called active sensing. Our approach to active sensing is to optimize the task plan (i) by minimizing an objective function, such as the expected execution time, which is an important criterion in industrial applications; and (ii) by constraining the task plans to plans which observe the geometrical parameter estimates to the required accuracy. Active sensing for compliant motion is a new research area. Hence, this paper primarily aims at formulating the active sensing problem and decoupling it into smaller optimization problems. The main contributions of this paper are (i) the definition of the CF-observable parameter space, which allows us to decouple the active sensing requirement for the task plan into a requirement for the CF sequence and requirements for the active sensing motions in each CF, and (ii) the description of practical (sub-optimal) solutions and heuristics, making online re-planning feasible.