Autonomous Vehicle Trajectory Planning With Probabilistic Cost-to-Go Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for determining trajectories in autonomous vehicles often result in inaccurate or suboptimal path selection due to limited time horizon-based cost evaluations, failing to consider long-term navigation impacts.
Innovation Solution
A planning system generates a local probabilistic graph that extends beyond the immediate time horizon, incorporating costs associated with navigating to a destination, including states, edges, actions, and probabilities, to determine optimal trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional time horizon-based cost evaluation techniques are used for trajectory determination, then the computational process is simple and fast, but the trajectory selection accuracy is poor and suboptimal paths are selected
Solution Approach 1:
The patent segments the trajectory evaluation into two distinct components: (1) immediate cost evaluation within the time horizon using conventional techniques, and (2) terminal cost-to-go evaluation beyond the time horizon using probabilistic graphs. This segmentation allows the system to maintain computational efficiency for immediate decisions while incorporating long-term strategic planning through the probabilistic graph component, thereby improving overall trajectory selection accuracy without overwhelming computational complexity
Solution Approach 2:
The patent performs preliminary action by pre-computing cost-to-go values for all possible states in the probabilistic graph before actual trajectory determination is needed. These pre-computed values represent the optimal cost from each state to the destination, considering the full future horizon. When a trajectory needs to be selected, the system simply retrieves these pre-computed values and combines them with immediate costs, avoiding the need for complex real-time optimization beyond the time horizon and improving selection accuracy efficiently
2Productivity
If conventional cost evaluation techniques are used, then the computational speed is fast, but the navigation efficiency and safety are reduced due to suboptimal trajectory selection
Solution Approach 1:
The patent introduces an intermediary component - the probabilistic graph with pre-computed cost-to-go values - that mediates between the simple immediate cost evaluation and the complex long-term optimization. This intermediary provides strategically optimal cost information from future states, which is combined with immediate costs to produce globally optimal trajectory selections. The intermediary enables the system to achieve high navigation efficiency through fast retrieval operations while ensuring trajectory optimality through the incorporation of long-term strategic costs
3Measurement precision
If the evaluation horizon is extended beyond immediate time horizon, then the trajectory optimality improves, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-computing all cost-to-go values for every state in the probabilistic graph offline or in advance. This pre-computation captures the full future horizon optimization without requiring real-time computation when trajectories need to be selected. During actual operation, the system only performs simple lookups and additions, achieving both high cost evaluation accuracy (by considering the full horizon) and fast processing times (by avoiding complex real-time optimization)
Solution Approach 2:
The patent makes the evaluation approach dynamic by adapting the computation depth to the situation: for immediate cost evaluation, it uses simple fast computations within the time horizon; for long-term strategic costs, it uses pre-computed values from the probabilistic graph. This dynamic approach allows the system to achieve accurate full-horizon evaluation accuracy while maintaining fast processing speeds by avoiding redundant computations
Data Source
AI summary
Techniques for determining a driving trajectory for an autonomous vehicle to follow are described herein. A vehicle may receive a destination associated with a region of an environment to which the vehicle is to navigate. Based on the destination, the vehicle can generate a local probabilistic graph that includes states, edges connecting the states, actions for the vehicle to perform along the graph, and/or probabilities associated with the actions. The vehicle may determine cost-to-go values for each state within the local probabilistic graph. While navigating to the destination, the vehicle can generate candidate trajectories. When determining the cost of following a candidate trajectory, the vehicle can determine the cost by projecting an ending state of the candidate trajectory onto the local probabilistic graph to determine an optimal action which considers a larger understanding of the goal of the vehicle. The vehicle can be controlled based on the candidate trajectory.