A mobile robot becomes more intelligent as its control system is given more capabilities to respond to its environment autonomously. This thesis develops a distributed real-time control system for a mobile robot which is intended to operate autonomously in an industrial environment. It is a unified approach to real-time sensing, planning, and control based on a parallel processing architecture.To be fully autonomous, a mobile robot must be able to sense its environment, build or update maps, plan and execute actions, and adapt its behaviour to environmental changes. The ability of a control system to support these complex tasks in real time is significantly affected by the organisation of information pathways within the architecture. After examining different architectures described in the literature, a transputer-based architecture is developed to maximise the parallel information flow from sensing to action to provide minimal delay in responding to a dynamically changing environment.Taking account of uncertainty, planning an optimal path is difficult since the internal world model quickly becomes invalid. To find a solution, dynamic changes in the environment are classified as different types of obstacles that are assumed to appear randomly. Bayes' theorem is applied to build statistical models to estimate the mean number unexpected obstacles encountered. This provides a feasible way for the global path planner to update its internal world model dynamically based on available sensor information. A dynamic programming algorithm is used to plan an optimal path.An array of sonar sensors is used to detect dynamic changes in the environment. To reduce uncertainty in noisy data, a probabilistic sensor model and rule-based heuristics are built. The collision avoidance problem is formulated using decision theory to achieve both collision-free and optimal solution. An optimal decision rule to avoid unexpected obstacles is calculated to minimise the Bayes risk in trading between a careful maneuver and an alternative ...