Abstract-We consider the problem of predicting the probability of an accident in working environments where human operators and robotic manipulators co-operate. We show how, starting from a stochastic discrete time system describing human motion, it is possible to construct a discrete abstraction of the system (a discrete time Markov Chain) to predict the possible trajectories starting from an initial point. The DTMC is used to predict the future evolution for the system, for a fixed horizon, pinpointing the states that, at each step, can be marked as dangerous. This way, the system estimates the probability of an accident and stops the robot when the result is greater than a threshold.

Mobile robots are increasingly utilised in automated plants to the purpose of moving wares and material between the different production lines and logistic areas. In this context, the presence of human operators in the facility is frequently allowed to carry out or supervise some phases of the production. The problem arises of how to make the coexistence possible with controlled risks for the operator and without affecting the productivity with frequent interruptions. In this paper we propose a solution to this problem based on a probabilistic technique. A system of visual sensor (mounted on the mobile robots) detects the presence of a human operator and a discrete abstraction (essentially a discrete-time Markov chain) is used to predict his/her motion and hence find the probability of an accidental injury. For the computation of the latter, we combine the probability of having a collision with a given speed with the probability of receiving an injury out of the collision (taken from physiological models suggested by the automotive literature).