Research On The Evaluation Method For Road Condition And The Control Strategy For Autonomous Obstacle Avoidance Applying To Small Intelligent Vehicle

With the popularization of the small intelligent vehicles, we have more options to the short-distance travel. In addition, the intelligent vehicle is gradually playing a more important role in some special occasions, such as the perform assistance to counter-terrorism, inspections and other special tasks, and thus the auxiliary driving technology to obstacle avoidance for the narrow and complex road have attracted a large of researchers. Throughout the domestic and foreign researches on intelligent vehicles, which were mostly about the driverless car. The self-driving cars have the high cost relying on multi-sensor technology and have the application limitation which mainly drive on structural roads, which caused that the core technology is not entirely applicable to small intelligent vehicles. In addition, researches on the control method of autonomous obstacle avoidance are roughly divided into the traditional algorithms adapted to the static condition and the intelligent algorithms adapted to the dynamic scenes. The avoidance decisions made by these methods for the narrow complex roads whose surroundings change quickly have not completely achieved the real-time autonomous obstacle avoidance effect, which have certain deficiencies.To solve these problems, an evaluation model based on BA-BP algorithm and an improved HSIC method based on Reinforcement Learning algorithm to autonomous obstacle avoidance for the small intelligent vehicle driving on unknown complex roads were researched in this thesis, aiming for accurate monitoring to the complex road which the intelligent vehicle is currently traveling on, extracting the reasonable obstacle information, improving the accuracy of obstacle avoidance, to provide the basis for driving assistance. The main topics of this thesis are as follows.1. Analysis and extraction to the five characteristics of the road condition evaluation. The image processing technology was introduced into this article to detect the road information, and five features which include the pavement roughness, the pavement curvature, the aspect ratio of the obstacle, the effective area ratio of the obstacle and barrier coefficient were extracted to reflect the current road condition. And on the basis, the definition and quantization of above characteristic parameters lay the foundation for the subsequent evaluation of road condition.2. An evaluation method of road condition based on BA-BP algorithm was researched. BP neural network was selected as the basis of evaluation model for road condition, and the bat algorithm was introduced to search better initial weights and thresholds of BP network in this thesis, to fill the gaps of random selection. In order to get more reasonable and efficient training process, adding the adjustment factor to highlight the major effect of the obstacles to road traffic. With five characteristics of road conditions as the input, the evaluation results were finally obtained by this evaluation model. We can get a more convenient and accurate description of the association relationship between the characteristic parameters and the road condition by this method, and the reasonable evaluation results can be obtained.3. In order to improve the control accuracy and real-time performance of autonomous obstacle avoidance, the autonomous obstacle avoidance method based on the reinforcement learning and human-simulated intelligent control algorithm was researched. The human-simulated intelligent control method was introduced by imitating human control strategy, while referring to the control algorithms of mobile robot, then the reinforcement learning was added to improve the HSIC algorithm. The ideal trajectory of obstacle avoidance was designed, and the avoidance parameters were built which contain acceleration and turning radius. Lots of learning correction resulted according to the continuous deviation information, and some parameters of the controller can be adjusted in real time by reinforcement learning. Accurate obstacle avoidance can be intuitively and quickly achieved by this method.The simulation experiment results show that the road condition evaluation accuracy of the evaluation model based on BA-BP algorithm reached 95.15%, which means that the unknown road information can reasonably be extracted and evaluated. In addition, the autonomous obstacle avoidance method for small road segment has good applicability, which can achieve the autonomous obstacle avoidance result. And the obstacle avoidance accuracy rate reached 92.86%, which has better effect compared with other methods. This autonomous obstacle avoidance method can satisfy the obstacle avoidance needs of small intelligent vehicle on the unknown roads, simultaneously with rapid response, high accuracy, etc., which have important theoretical and practical significance to the research on driving assistance technology.