International audience

Large-scale natural and human-caused disasters have created significant challenges for worldwide Search and Rescue (SAR) operations, highlighting persisting concerns related to the efficiency and technical limitations of existing technologies. To address these challenges, the proposed Universal Platform for Search and Rescue integrates various technologies, including a voice-guided control system, advanced 3D reconstruction techniques, and a people tracker and follower system. A central feature of our work is the platform’s universality: our system acts as an additional, modular controller that can connect to any robotic platform—commercial or custom—that supports text-based command communication via network or cable. The system does not replace original robot logic, but rather extends capabilities with minimal integration. Tests showed that the platform can effectively execute voice commands and track a specified route even in high-wind (23 km/h) and noisy environments (70–100 dB for the Drone, 65–99.6 dB for the Quadruped), providing a user-friendly and intuitive interaction for users across different skill levels. Performance metrics indicated strong quality in 3D scene reconstruction with significant similarity between the reconstructed images and reference images (Drone: indoor: 0.82 SSIM, outdoor: 0.81 SSIM; Quadruped: indoor: 0.79 SSIM, outdoor: 0.58 SSIM). Consequently, the immersive 3D mapping reconstruction facilitated prompt and precise terrain assessments for both internal and external operations. Furthermore, the integration of real-time video streaming and cloud-based connectivity optimized the data flow and strengthened communication during operations, allowing person face identification, 3D tracking, and following.

Robots are not yet ready to handle unstructured and uncertain environments by adapting their trajectory online depending on the changes in the surroundings. Knowing how to handle moving objects is an essential skill for robots in many industrial and service robotics applications. This work presents a reactive grasp controller inspired by an established human motion model. It starts planning by generating a straight line from the current pose of the end-effector to the target object pose. Then, the line is shaped considering the new target position. A trace controller guarantees that the end-effector remains on the path, and the collision-aware velocity regulation feature ensures to safely proceed on it by modulating the advancing velocity until completely stopping the motion if an obstacle is encountered on the path toward the target pose goal. Control actions are sent to the robot controller as smooth velocity commands at 1 KHz to be reactive. The capabilities of the proposed planner are showcased on a real robot in three dynamic experimental settings relevant to industry: (i) grasping objects moving on a conveyor belt, and two human–robot collaboration tasks that are (ii) grasping objects that can move during the robot’s motion toward the target due to the interaction of a human operation, and (iii) human-to-robot handover. A comparison with state-of-the-art methods is also discussed.