Stereo Vision-Based Spill-Free Serving Robots With Critical Acceleration Awareness

This paper presents an adaptive online acceleration control system for serving robots that can transport open liquid containers without spilling by leveraging stereo vision perception models. Although previous models successfully suppressed sloshing motion by reducing inertial forces acting on the robot, liquids often spilled because they disregarded the effects of container size and liquid levels. Furthermore, a sloshing simulation is computationally heavy, and thus real-time prediction is very challenging. To solve these issues, we propose a stereo vision-based parameter recognition system that can detect liquid levels and container dimensions even when the liquid is transparent. The proposed system can also monitor multiple containers simultaneously by rapidly estimating critical accelerations without heavy computation. Here, critical acceleration refers to the threshold beyond which spilling occurs, depending on the container size and liquid level. Spill-free velocity profiles of the robot can be generated in real-time. The spill-free delivery ability was verified by experiments with varying container size and liquid volumes. The system maintained stable motion and prevented spillage. Moreover, the serving robot can also stop without spillage when a moving obstacle suddenly obstructs its trajectory. Our approach enhances safety and efficiency for liquid delivery for indoor service system.