Research On Multi-objective Dynamic Obstacle Avoidance Method Based On Efficient Grid Map Evaluation

Path planning in dynamic environment is an important guarantee for the robot to complete the navigation task safely and efficiently,especially in the environment with pedestrians,the robot must have certain basic interaction reaction.Most of the existing algorithms set up a special object detection module to obtain pedestrian information,and even equipped with a proprietary sensor to track and predict the movement of pedestrians.However,there are two problems: one is the extra computation cost in object recognition,tracking,prediction and multi-source sensor information fusion,which greatly reduces the real-time performance of the follow-up obstacle avoidance system and poses a threat to human life directly;the other is the unpredictability of human actions.As a classical holonomic constraint object,human beings have the ability to change their velocity and direction at any time,at any place.The prediction based on the prior behavior of pedestrians is not reliable in a strict sense.In order to improve the reliability and safety of the robot obstacle avoidance algorithm in the dynamic environment where pedestrians participate,inspired by the selfdriving technology,this paper introduces the dynamic grid map which has more advantages in the dynamic environment representation and multi-source sensor information fusion,and studies how to efficiently calculate the dynamic grid map and how to use the velocity of the dynamic objects.A multi-objective dynamic obstacle avoidance method is proposed:Firstly,to perceive the dynamic environment,it is defined as a random dynamic system.Design a multi-objective state estimator PHD/MIB filter.The D-S evidence theory is introduced to approximate the unobserved grid cells,reduces the calculation for unnecessary environment representation.Complete the particle realization of DSPHD/MIB filter,and the filtering is validated with our simulation.Secondly,to improve the computing speed of particle filter and meet the demand of the path planning algorithm for environmental information,based on the GPU parallel acceleration algorithm of traditional grid map.The particle sorting step is added to simplify the correlation calculation between particles and grid cells.The average single frame data processing cost is compressed to 66 ms.The computing efficiency is increased by 22.87%.Furthermore,according to the task characteristics of robot avoiding pedestrians,design the critical and the active region under the pedestrian model as the virtual force critical condition of QVFF method.The relative velocity is taken as the kinematic interpretation of the "detour force" in the original algorithm,and establish a new detour force function,which effectively reduces the response time of the robot avoids pedestrian,increases the minimum distance between the robot and the pedestrian model by 13.9% ? 38.36%.In another word,it improves the system security.Finally,with the help of the software structure of layered cost map,an information connection bridge is built between the perceptual system and the path planning algorithm.The corresponding local cost map generated by the dynamic grid map is embedded to the ROS navigation stack,completing the construction of simulation experiment framework,achieving compatibility with the mainstream open source path planning algorithm.