Off-road robotics - perception and navigation

The research in off-road robotics is focusing on autonomous or semi-autonomous vehicles able to navigate in rough terrain without a predefined trail. Typical examples are agriculture and forestry machines or planetary rovers. Detailed environmental perception and navigation strategies under various soil conditions have to be examined for developing an adequate control concept for such vehicles. Due to noisy, erroneous sensor measurements and rough ground conditions, it is essential, that the control architectures are characterized by a high robustness. The often used sense, plan, act strategies have the drawback that in case of unexpected states, re-planning of control sequences has to be determined. For the realization of a behavior-based control architecture for off-road vehicles, as those developed in the RRLab, it is essential to have a well-defined design strategy. In our case a bottom-up strategy is applied, which derives percepts starting from the senor system and basic control behaviors related to the active DOF of the vehicle. Typical basic control behaviors, which are mainly reactive, are safety (stop in front of obstacles), collision avoidance (drive around obstacles) or targeting (reaching a specific pose in the operational environment of the vehicle). For deliberative control the behavior, input is replaced by specific maps or scene descriptions. The often generated sequence of actions can be mapped to a specific design pattern of the iB2C network or can be used to stimulate several related behaviors in parallel.