Autonomous Vehicle Path Planning With Offline Graphs for Dynamic Obstacles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current path planning algorithms for vehicles in dynamic environments are limited by the computational expense of generating large graphs, leading to suboptimal solutions and abnormal system behavior.
Innovation Solution
The method involves generating an offline graph representing feasible transitions in the environment, annotating it with the current position of the vehicle, and determining a path through a non-exhaustive search, allowing for efficient adaptation to new information and long-run goals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large graphs are generated to represent feasible paths in dynamic environments, then path planning completeness and optimality are improved, but computational time and processing resources are excessively increased
Solution Approach 1:
The patent segments the path planning process into two distinct phases: an offline phase where a comprehensive graph is pre-computed, and an online phase where only minimal graph updates are performed. This segmentation allows the computationally intensive graph generation to be done beforehand, while runtime operations only require lightweight updates, thus resolving the contradiction between complete path planning and computational time.
Solution Approach 2:
The patent performs preliminary action by pre-computing the complete path planning graph offline before the vehicle operates in the dynamic environment. This preliminary computation ensures that all feasible paths are identified in advance, and during online operation, the system only needs to annotate the pre-existing graph with current position information, avoiding the need to regenerate the entire graph and thus significantly reducing computational time.
2Productivity
If small and shallow graphs are created to reduce computational burden, then processing speed is improved, but path planning algorithms fail to converge to near-optimal solutions
Solution Approach 1:
The patent segments the graph into a comprehensive offline component that ensures optimality and a minimal online annotation component that ensures speed. The offline graph contains all feasible paths and is computed once, guaranteeing near-optimal solutions, while the online phase only adds current position annotations, maintaining processing speed without sacrificing solution quality.
Solution Approach 2:
By performing the comprehensive graph generation as a preliminary offline action, the patent ensures that all optimal paths are identified before runtime. During online operation, only minimal annotations are added, which maintains processing speed while the pre-computed graph structure ensures that near-optimal solutions are always available.
3Adaptability or versatility
If comprehensive graphs are generated to capture all feasible paths, then adaptability to dynamic obstacles is improved, but memory allocation and computational resources are excessively increased
Solution Approach 1:
The patent segments the graph data structure into a permanent offline graph stored in persistent memory and a temporary online annotation layer. The comprehensive graph structure is built once offline and stored, while runtime adaptability to dynamic obstacles is achieved by adding minimal annotation data about current positions and obstacles. This segmentation allows comprehensive path coverage without proportionally increasing runtime memory allocation.
Solution Approach 2:
The patent performs preliminary computation of the complete graph structure offline and stores it in persistent memory. During online operation, the system only needs to allocate minimal memory for annotating current positions and obstacle information, rather than allocating memory for the entire comprehensive graph structure. This preliminary action significantly reduces runtime memory requirements while maintaining adaptability.
Data Source
AI summary
Techniques for traversing in an environment that includes at least one obstacle, by a mobile autonomous system, to a destination in the environment, are presented. The techniques can include generating, prior to the mobile autonomous system commencing activity in the environment, a graph including a plurality of vertices representing positions in the environment and a plurality of edges between vertices representing feasible transitions by the mobile autonomous vehicle in the environment; annotating the graph with at least one edge connecting a representation of a present position of the mobile autonomous system to a vertex of the graph; determining, based on the graph, a path from the present position of the mobile autonomous system in the environment to the destination; and traversing the environment to the destination, by the mobile autonomous system, based on the path.