Fleet Traffic Planning With Search Tree Pruning for Deadlock Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing centralized vehicle control systems face computational challenges in managing a scalable number of autonomous vehicles, particularly in scenarios with limited computational resources, leading to inefficiencies and potential deadlock states.
Innovation Solution
A traffic planning method that generates a search tree with a tree-expansion criterion to limit computational load by discarding non-promising trajectories, ensuring vehicles move from connected and unoccupied resources, and using a discount function to evaluate and propagate rewards, thereby optimizing resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized control is used to coordinate multiple autonomous vehicles, then productivity and resource utilization are improved, but computational complexity and resource consumption increase significantly
Solution Approach 1:
The patent segments the computational problem by dividing the fleet coordination task into individual vehicle trajectory planning problems. Each vehicle independently plans its trajectory based on shared resource availability information, rather than computing coordinated trajectories for all vehicles simultaneously. This segmentation reduces computational complexity from exponential (considering all vehicle combinations) to linear (considering each vehicle separately), while maintaining productivity through centralized resource coordination.
2Reliability
If the planning horizon is extended to improve foresight and avoid deadlocks, then reliability is improved, but computational time and resource consumption increase
Solution Approach 1:
The patent applies preliminary action by having vehicles plan their trajectories over an extended planning horizon that includes future time steps beyond immediate needs. This allows vehicles to anticipate potential deadlock situations and avoid them proactively by selecting alternative trajectories. The computational time is managed by using efficient search algorithms that can handle the extended horizon without exhaustive exploration, balancing foresight with computational constraints.
3Measurement precision
If the search tree breadth and depth are increased to evaluate more trajectories, then measurement precision is improved, but the computational budget is exceeded
Solution Approach 1:
The patent changes key parameters of the search process: it limits the maximum depth of the search tree to a predefined horizon and restricts the breadth by considering only relevant trajectories that respect resource constraints. The evaluation accuracy is maintained by using a scoring function that efficiently ranks trajectories based on multiple criteria (resource utilization, productivity, deadlock avoidance). This parameter optimization allows sufficient trajectory evaluation without exceeding computational budget.
4Productivity
If more vehicles are added to the fleet to increase productivity, then output is improved, but computational resources required for centralized control increase
Solution Approach 1:
The patent implements universality by designing a standardized trajectory planning algorithm that can handle any number of vehicles using the same computational framework. Each vehicle uses the same planning logic and resource constraint checks, allowing the system to scale to larger fleets without proportionally increasing control system complexity. The centralized resource coordination mechanism remains unchanged regardless of fleet size, enabling scalable productivity improvement.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented traffic planning method (100) for controlling a plurality of vehicles which are movable among multiple shared resources, comprising: receiving (no) a transport mission; generating (112) a root node representing an initial resource occupancy of the vehicles; sequentially generating (114) a search tree from the root node, in which each edge represents a motion command and each node represents a resource occupancy, wherein each node is associated with a score including an anticipated reward for fulfilling the transport mission; identifying (116) a target node with an acceptable score; and deriving (118) a planned sequence of motion commands corresponding to a path to the target node, wherein: child nodes of a leaf node are generated (114.4) subject to a tree-expansion criterion (114.2); a reward is assigned (114.6) to a child node if the transport mission is fulfilled; and an anticipated value of the assigned reward is added (114.8) to all parent nodes.