A generic Bayesian perception framework, designed to estimate a dense representation of dynamic environments, by fusing and filtering multi-sensor data, has been developed, implemented and tested on embedded devices. The main features of the approach are the followings: • Data from multiple sensors are properly fused in probabilistic occupancy grids. • Motion and robust occupancy are estimated by a specific Bayesian filter. • Short-term collision risks and object parameters are assessed. • The whole system has been implemented and tested on Nvidia embedded devices, and produces real-time results.