Optimization of Dynamic Obstacle Avoidance Algorithms for Autonomous Mobile Robots on Uneven Terrain

The paper studies the solution of the dynamic obstacle avoidance problem of autonomous mobile robots operating in uneven terrain environments. Wheel slip, inclination effect, and increase in kinetic energy on rough surfaces reduce the effectiveness of the traditional Artificial Potential Field (SPS) method, especially when the obstacles move nonlinearly, which leads to the problem of local minimum. To overcome this shortcoming, the parameters of the SPS method (attraction and repulsion coefficients, radius of influence, step coefficient) are optimized using Genetic Algorithm (GA). The kinematic and dynamic model of the robot is extended with an inclination-dependent deceleration function, and a LiDAR/camera-based sensor model is used for real-time detection of dynamic obstacles. Experiments conducted in a series of 10 independent trials in a Python-based simulation environment showed that the proposed SPS-GA method reduced the target arrival time by an average of 31.4%, the trajectory length by 5.63%, and an increase in the energy indicator by 2.25% (the result of moving at speed) compared to the classical SPS method. The success rate for both methods was 10/10. The results show that GA-based optimization is a flexible tool that allows adjusting the speed-energy balance in dynamic environments.