Optimal design of reflective sensors using probabilistic analysis

Linear stepper, or Sawyer, motors have become popular in robotic mechanisms because of their high positional accuracy in open loop control mode [1]. Precision and repeatability are prerequisites in manufacturing and assembly. However the motor’s actual position becomes uncertain when it is subject to external forces. Position sensors mounted on the motors can solve this problem and provide for force-control [2].This paper describes a sensor, a technique for determining the robot’s position, and an analysis technique for determining the optimal sensor configuration. Reflective optical sensors are used to generate raw data which is scaled and then processed using Bayesian probability methods. We had wanted to analyze different sensor configurations by marginalizing the performance over predicted data. Since marginalizing over the entire state space is infeasible due to its size, Monte Carlo techniques are used to approximate the result of the marginalization. We implemented the positional technique, and measured its performance experimentally; the sensors estimated the robot’s position to within 2/1000 ″, in line with the probabilistic analysis.