Autonomous Vehicle Path Planning With Predicted Traffic Reactions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional path planning for autonomous vehicles does not adequately account for the probable reactions of nearby vehicles, leading to sub-optimal navigation paths and potential delays or failures in achieving destinations.
Innovation Solution
Integrate a decision tree-based approach that predicts the probable reactions of nearby vehicles using machine learning, considering their behavior and interactions with the autonomous vehicle, and utilizes a value function to determine the highest value navigation path, incorporating deep learning for optimization and data reduction techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional path planning approaches are used, then navigation planning can be performed in many circumstances, but the path planning accuracy deteriorates because nearby vehicle reactions are not accounted for
Solution Approach 1:
The patent merges the prediction function and path planning function into a single integrated neural network system. The prediction network predicts nearby vehicle reactions while the planning network simultaneously determines optimal paths, allowing the system to account for vehicle interactions without requiring separate complex modules. This integration resolves the contradiction by achieving high path planning accuracy through a unified architecture that inherently considers nearby vehicle behaviors.
Solution Approach 2:
The patent introduces an intermediate representation of nearby vehicle reactions as a mediator between the input sensor data and the final path planning output. The prediction network generates predicted reactions of nearby vehicles that serve as intermediate inputs to the planning network, enabling accurate path planning while maintaining system organization. This intermediary layer resolves the contradiction by structuring the complex system into manageable functional stages.
2Reliability
If predictions of nearby vehicle reactions are incorporated into path planning, then path accuracy improves, but the decision time increases
Solution Approach 1:
The patent performs preliminary prediction of nearby vehicle reactions using the prediction network before the planning network determines the final path. By pre-computing the likely reactions of nearby vehicles and incorporating them as inputs to the planning stage, the system avoids iterative simulations during real-time decision making. This preliminary action resolves the contradiction by preparing necessary information in advance, enabling both accurate path planning and fast decision execution.
Solution Approach 2:
The patent segments the decision-making process into two distinct but integrated neural network components: a prediction network that specializes in forecasting nearby vehicle reactions, and a planning network that specializes in determining optimal paths. This segmentation allows each network to be optimized for its specific function and to operate efficiently in parallel or sequence, resolving the contradiction by dividing the computationally intensive task into specialized sub-tasks that can be executed faster than a monolithic system.
3Productivity
If conventional path planning is used, then the system remains simple, but sub-optimal paths are generated when nearby vehicles have certain reactions
Solution Approach 1:
The patent creates a universal neural network system that simultaneously performs multiple functions: sensing the environment, predicting nearby vehicle reactions, planning optimal paths, and adapting to different traffic scenarios. The integrated architecture handles diverse navigation situations (intersections, merges, overtaking) through a single unified model rather than separate specialized systems. This universality resolves the contradiction by achieving high navigation efficiency across multiple scenarios through one multi-functional system that learns optimal behaviors during training.
Solution Approach 2:
The patent utilizes parameter changes in the neural network weights and configurations to adapt the system's behavior to different traffic conditions and scenarios. During training, the network learns optimal parameters for various situations, and during operation, it selects or adjusts parameters based on the current context. This parameter adaptability allows the system to achieve high navigation efficiency in diverse scenarios without requiring fundamentally different architectures for each case, resolving the contradiction between performance and complexity.
Data Source
AI summary
Sensors measure information about actors or other objects near an object. Sensor data is used to determine a sequence of possible actions for the maneuverable object to achieve a determined goal. For each possible action to be considered, one or more probable reactions of the nearby actors or objects are determined. This can take the form of a decision tree in some embodiments, with alternative levels of nodes corresponding to possible actions of the present object and probable reactive actions of one or more other vehicles or actors. Machine learning can be used to determine the probabilities, as well as to project out the options along the paths of the decision tree including the sequences. A value function is used to generate a value for each considered sequence, or path, and a path having a highest value is selected for use in determining how to navigate the object.