Robots developed from the 60’s to the present have been restricted to highly structured environments such as work cells or automated guided vehicles, primarily to avoid harmful interactions with humans. Next generation robots must function in unstructured environments. Such robots must be fault tolerant to sensor and manipulator failures, scalable in number of agents, and adaptable to different robotic base platforms. Layered Mode Selection Logic (LMSL), a robot-controller architecture, addresses these concerns. The LMSL architecture has been implemented and tested on UALR’s J5 robotics research platform. Objects are classified by acceleration and force measurements. Collisions are detected, and objects are classified by how difficult they are to push. J5 either avoids the obstacles or manipulates them depending on this classification. Comparable results are achieved with all sensors functioning, with only the acceleration sensor (force sensor faulted), and with only the force sensor (acceleration sensor faulted). A second demonstration of the architecture addresses the problem of border security. A team of autonomous robots are configured as a flexible sensor array. A one-dimensional experiment using Lego robots shows that the robots distribute themselves evenly along the border until an intruder penetrates the border. The robots at the point of penetration cluster, whereas the robots removed from the point of penetration remain evenly distributed, albeit with a greater inter-robot separation.