Obstacle Avoidance Control for Robotic Manipulators in Unstructured Environments

Obstacle avoidance is one of the core competencies of robotic manipulators that will be performing operations in an unstructured environment, with dynamic situations, incomplete perception, and variable distributions of obstacles. Unlike other organized industrial contexts, this type of environment requires robots to perceive, interpret, and act on uncertainty in real time and, at the same time, maintain the safety and efficiency of their tasks. In this paper, a detailed research of obstacle avoidance systems of control that can be applied in the robot manipulator is carried out in unstructured environments. The existing methods can generally be categorized into reactive sensor methods and model-based pre-planning techniques. Reactive approaches, such as artificial potential fields and locally sensor-based control, have low-Latency reaction but lack awareness of the global position and local minima. Model-based methods, in contrast, build a representation of the environment, such as a point cloud or a voxel map, and solve obstacle avoidance by either an optimization or a sampling-based planning problem, thus naturally solving complicated constraints and global goals. Recent studies in learning-based planning, which include reinforcement learning, imitation learning, and the vision-language-action (VLA) models, are reviewed. Especially regarding their ability to generalize test environments into real cases. Lastly, the strengths and weaknesses of different obstacle-avoidance paradigms are then discussed, and the future research directions towards strong, scalable, and intelligent manipulation in unstructured environments are outlined.