Autonomous tactile exploration and mapping with a redundant robotic arm

Research in construction robotics is a rapidly growing field, driven by the need to automate operations carried out in unstructured and difficult-to-standardize environments. This thesis is set in a construction scenario involving two agents: a crane and a robotic arm. The system assigns the crane the task of coarse positioning of masonry blocks, while the robotic arm is responsible for their precise placement. Achieving the required accuracy demands a preliminary assessment of the position and orientation of each block through an exploration and mapping process, for which this thesis proposes a complete solution based on an RGB-D camera system and a robotic arm operating via physical contact. Visual perception alone proves insufficient to meet the accuracy requirements of fine placement, but remains useful for a preliminary scan of the scene, carried out through point cloud filtering and a RANSAC algorithm for geometric plane detection. A 7-DOF redundant Franka Panda manipulator, equipped with a flat circular end-effector, physically probes the wall surfaces. A hybrid controller guarantees planar alignment and compliance with the nominal force limits, exploiting the intrinsic accuracy of the robot’s encoders to measure the position and orientation of each face. An OMPL RRT planner produces safe trajectories, avoiding collisions between the robot links and with the environment. Experimental validation confirms high robustness to the limited accuracy of the stereo cameras, with orientation errors below 1.5◦ on four out of five planned targets.