Autonomous Vehicle Action Planning via Tree Search Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional action planning systems for autonomous vehicles, based on finite state machine approaches, only allow passive reaction to environmental changes and lack a future look-ahead, failing to effectively evaluate how actions might impact future traffic situations.
Innovation Solution
A computer program and method that uses a tree search through possible action sequences to estimate the future environment of an autonomous vehicle, predict the motion and reactions of dynamic obstacles, and decouple action decision time resolution from iterative prediction resolution, enabling the generation of trajectories and interactions with other vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a finite state machine approach is used for action planning, then the system implementation is simple, but the vehicle can only passively react to environmental changes without future look-ahead capability
Solution Approach 1:
The system transitions from a static finite state machine to a dynamic planning approach where the vehicle actively simulates future scenarios. The action planning module dynamically generates multiple possible action sequences and evaluates their outcomes, enabling the vehicle to adapt to future traffic situations rather than merely reacting to current conditions.
Solution Approach 2:
The system performs preliminary evaluation of multiple possible action sequences before executing any action. By simulating future traffic scenarios and predicting outcomes of different action combinations, the system prepares ahead of time to select the optimal action, rather than reacting passively when situations arise.
2Adaptability or versatility
If the system evaluates multiple action sequences with future look-ahead, then the decision-making intelligence improves, but the computational complexity increases
Solution Approach 1:
The action planning process is segmented into discrete time steps and action sequences. The system divides the planning horizon into manageable segments, evaluating actions at specific time intervals rather than continuously. This segmentation reduces computational complexity while maintaining decision-making intelligence by focusing evaluation on critical decision points.
Solution Approach 2:
The system evaluates a limited number of plausible action sequences rather than all possible actions. By focusing computational resources on the most relevant action combinations based on current traffic context, the system achieves sufficient decision-making intelligence without the full computational burden of exhaustive evaluation.
3Measurement precision
If the system uses iterative prediction over timesteps, then the prediction accuracy improves, but the computation time increases
Solution Approach 1:
The system performs iterative prediction at periodic time steps rather than continuously. By updating predictions at discrete intervals and using the previous prediction results as initial conditions, the system maintains prediction accuracy while reducing computation time compared to continuous prediction.
Solution Approach 2:
The iterative prediction process maintains continuity by using results from previous timesteps as the basis for current predictions. This allows the system to build upon existing predictions rather than starting from scratch at each timestep, improving accuracy while managing computation time efficiently.
Data Source
AI summary
An action planning system (100) and method for autonomous vehicles are provided. The system (100) comprises one or more processors (108) and one or more non-transitory computer-readable storage medium (110) having stored thereon a computer program used by the one or more processors (108), wherein the computer program causes the one or more processors (108) to estimate future environment of an autonomous vehicle (114), generate a possible trajectory for the autonomous vehicle (114), predict motion and reactions of each dynamic obstacle in the future environment of the autonomous vehicle (114) based on current local traffic context, and generate a prediction iteratively over timesteps.