Autonomous Vehicle Fleet Planning Under Fixed Compute Budgets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for centralized vehicle control, such as those described in US20200247611A1, inefficiently utilize computational resources, especially when dealing with a scalable number of vehicles, as they do not effectively leverage unused computational resources and can lead to increased computational burden with more vehicles.
Innovation Solution
The proposed method optimizes the use of a predefined computational budget by introducing preliminary steps to eliminate nonfeasible motion commands, allowing for a higher efficiency in computational resource usage and suggesting a suitable search depth reduction to offset the increase in computational burden when more vehicles are deployed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a preset maximum search tree depth is used for motion planning, then the computational process is bounded, but computational resources are not efficiently utilized when many states are not viable
Solution Approach 1:
The patent applies preliminary action by performing feasibility checks and filtering non-viable states before conducting the full search tree expansion. This preliminary filtering removes states that cannot lead to valid solutions, ensuring that computational resources are focused only on promising paths and preventing waste on impossible configurations.
Solution Approach 2:
The patent dynamically adjusts search parameters such as search depth and branching factor based on the viability of states encountered during planning. When non-viable states are detected, the algorithm modifies search parameters to reduce exploration in those directions, thereby adapting computational resource allocation to the actual problem structure.
2Adaptability or versatility
If the number of vehicles is increased, then fleet coverage and capability are improved, but computational burden increases significantly
Solution Approach 1:
The patent segments the motion planning problem by decomposing it into individual vehicle sub-problems that can be solved independently or with limited coordination. Each vehicle's planning is handled separately through automated driving planners, while a higher-level coordinator manages interactions. This segmentation allows the system to scale to larger fleets without exponentially increasing computational complexity.
Solution Approach 2:
The patent applies partial action by focusing computational resources on planning horizons and state spaces that are sufficient for safe and efficient operation, rather than exhaustively exploring all possible states. The search tree depth and breadth are tuned to provide adequate planning coverage without unnecessary computational expenditure, enabling scalability to larger vehicle counts.
3Ease of operation
If a fixed computational budget is allocated, then resource management is simplified, but the system cannot adapt to varying planning requirements
Solution Approach 1:
The patent implements dynamic resource allocation where the computational budget for motion planning is not fixed but adapts based on the specific planning situation. Factors such as traffic density, vehicle interactions, and planning horizon requirements influence the allocated computational resources. This dynamic approach maintains ease of operation through automated resource management while providing the flexibility needed for varying planning scenarios.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A traffic planning method for controlling a plurality of vehicles (v1, v2, v3, v4) is disclosed, wherein each vehicle occupies one node in a shared set of planning nodes (wp1, wp2, wp3, wp4, wp5, wp6, wp7, wp8) and is movable to other nodes along predefined edges between pairs of the nodes in accordance with a finite set of motion commands. The method comprises: obtaining initial node occupancies of the vehicles; from said finite set of motion commands, determining a mean number of feasible motion commands in a neighborhood of the initial node occupancies; determining a search depth d which makes optimal use of a predefined computational budget; and determining a suitable sequence of motion commands by means of an optimization process which considers the search depth d.