Together with end-users, requirements for the CENTAURO disaster-response system have been defined. Evaluation tasks and performance metrics have been specified. From this, specifications for system components have been derived and concepts for robot design, operator interfaces, modelling & simulation, and autonomous navigation & manipulation have been devised.
The Centauro robot has been designed and manufactured. It has a centaur-like body plan with four legs and an anthropomorphic upper body. The robot is driven by newly developed high-performance torque-controlled series-elastic actuators. Its lower body possesses four articulated legs with five joints each, ending in steerable, actively driven wheels to allow for omnidirectional driving as well as stepping locomotion. The two robot arms have seven joints each and its multi- fingered hands have complementary capabilities. The robot is equipped with a multimodal sensor head consisting of a 3D laser-range finder with spherical field-of-view, panoramic color cameras, and a depth camera. Additional sensors for environment perception are placed at the robot wrist and its base. Real-time control software has been developed, which includes whole-body motion generation and dynamic balance. The robot can be powered by a rechargeable battery and a wireless data link enables untethered operation.
A bi-manual exoskeleton has been designed and manufactured, to capture the arm motion of the main operator and to provide force-feedback during teleoperation of the robot. The exoskeleton has seven joints per arm. Finger motion capture and force feedback is provides by a hand exoskeleton. Bilateral teleoperation control over delayed communication has been realized. Furthermore, a head-mounted display providing live immersive visualization of the robot environment has been integrated.
A simulatable model of the robot and its environment has been developed as a virtual test bed. This digital twin has been used to predict the outcome of specific maneuvers before the real robot executed the selected action.
Efficient software for 3D mapping and semantic terrain classification has been developed to assess locomotion costs. This served as basis for planning hybrid driving-stepping locomotion. Perception software for 3D reconstruction of the manipulation work space, object detection, semantic segmentation, and 6D object pose estimation has been developed. Methods to grasping unknown objects by transferring knowledge from other instances of the same object category have been developed. They are integrated with efficient manipulator trajectory optimization.
The developed disaster-response system has been evaluated according to the identified requirements and performance measures. In the final evaluation, the Centauro robot demonstrated a large variety of challenging locomotion and manipulation tasks.
During the project, 93 scientific publications and 97 other dissemination activities have been reported. Potential customers and stakeholders have been analyzed and suitable ways to approach them have been identified. First hardware components have been introduced to the market and multiple software components and data sets have been released open source. The results achieved in CENTAURO form a solid basis for already running and future research projects.
