Enhancing Indoor Auto-Steering for AMRs through RGB and Depth Fusion

This research presents a method for improving the navigation capabilities of autonomous mobile robots (AMRs) in indoor environments. Indoor navigation is challenging due to the presence of various obstacles such as floors, walls, furniture, and doors. While depth sensing devices can effectively recognize geometric conditions in corridor environments, they struggle with reflective surfaces and slim objects. Our proposed solution is to fuse depth and RGB inspections using a dual -ResNet architecture in the visual detection ConvNet. This improves performance compared to traditional depth -only approaches. Field tests have shown that our system operates at a speed of 30 frames per second and guides the AMR through various corridor routes at 2 m/s, all on an embedded PC.