Researchers Develop a New Navigation System for Flying Robots
This bio-inspired, vision-based obstacle avoidance system improves flying robot’s navigation in dynamic environments.
Researchers at Shanghai Jiaotong University have developed a bio-inspired and computer vision-based obstacle avoidance system that improves a flying robot’s navigation in dynamic environments. The system is inspired by an owl’s ability to detect and avoid objects or animals in their surroundings.
The team outfitted a quadrotor with a servo motor and stereo camera to imitate how owls move their eyes in different directions while detecting still and moving objects around them. In this design, the servo motor behaves similarly to a neck, and the stereo camera acts as a head. The lightweight stereo camera allows it to move more quickly than the robot’s body, and its movements barely affect the quality of the robot’s movements or the direction it’s maneuvering in.
The system utilizes a sensor-planning algorithm to predict how much the robot benefits from detecting objects in various directions and plans the angle at which the head rotates. Therefore, the quadrotor can continuously and actively sense its surroundings, identifying obstacles that may obstruct its path.
The system also monitors and predicts the trajectories of moving objects in its path, adjusting its movements to changes in the environment. Based on collected data from the stereo camera, the system utilizes a sampling-based path planner to prepare a collision-free path. This outlines a robot’s movements so that it could arrive at its destination without crashing into other objects and sustaining damage.
Experiments were performed in real environments where a quadrotor was expected to reach a specific location while avoiding obstacles or monitoring and capturing an artificial rat. The results show a lot of promise since the robot performed exceptionally well on both tasks. It was able to adapt to sudden changes in its environment and avoid collisions with still and moving obstacles.
In the future, this system could perform missions in different environments, which range from urban areas to wildlife inhabited environments. Other researchers could use the system as a source of inspiration to develop flying robots with improved obstacle avoidance capabilities. Next, the team plans on developing systems that replicate how other animals behave while utilizing reinforcement learning techniques to enhance their system’s sensing performance.