Validation of algorithms developed by assistance robotics research on real platforms is essential. Producing robotic architectures that promote scientific advances while regarding usability for the final user is a challenging issue where an appropriate trade-off between both requirements must be found. This paper proposes a new framework for the development of mobility assistance techniques to improve the quality of life of elder population using a robotic wheelchair. An example of improvement of the wheelchair navigation is presented. The control of the wheelchair is done using a Bayesian semi-autonomous approach that estimates the user intended destination from input commands given through a Kinect sensor. Safe and comfortable autonomous navigation is performed using a probabilistic navigation method that takes into account the dynamics of the environment and human social conventions.