Autonomous vehicles, such as automatic-driving cars, drones and unmanned aerial or underwater vehicles, are increasingly used in commercial and other missions (including ground, air, and underwater missions) to collaboratively perform specific tasks. There are numerous optimization problems in the planning and operation of these autonomous vehicles. To account for the uncertainty related to the autonomous vehicles and their dynamic environment, a probabilistic method to the analysis, modeling and optimization is needed and beneficial. However, probabilistic optimization in these problems is highly challenging due to the complexity of these probabilistic algorithms and the long time needed to find the probabilistic solution. This paper is presents several innovative strategies that can greatly accelerate the probabilistic optimization solution process for autonomous vehicles within a unified, probabilistic graph-based modeling and computational framework. The acceleration strategies consider hybrid analytical/simulation-based approach, deterministic and stochastic network reduction and decomposition, adaptive sampling reduction and computational parallelisms all together. The principles and implementation methods of these strategies are discussed in detail in this paper. These strategies are then used in a path planning problem considering stochastic costs. Representative results are provided to demonstrate the benefits, effectiveness, accuracy and sensitivity of the proposed strategies and methods. The optimal solution with the highest probability in stochastic optimization is found to be same as the optimization result obtained from deterministic optimization considering the expected mean values, but stochastic optimization provides more information such as the probability distribution of multiple possible optimal solutions instead of a single solution. This study suggests that, using these strategies, the time needed to find the solution can be reduced by one to three orders of magnitude, compared with the traditional Monte-Carlo simulation method.
Accelerating Probabilistic Optimization Solution to Autonomous Vehicles under Uncertain and Dynamic Environments
2018-07-01
2549610 byte
Conference paper
Electronic Resource
English
Navigating autonomous vehicles in uncertain environments with distributional reinforcement learning
SAGE Publications | 2024
|