Actor Occupancy Corridor Prediction With Reachability and AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art actor prediction models for autonomous vehicles are either not robust enough to cover the space of possible future positions of actors on the road or are overly conservative, particularly in complex and chaotic driving environments, leading to inefficiencies in maneuvering.
Innovation Solution
A hybrid data-driven method using reachable sets and supervised learning to predict actor occupancy corridors, incorporating reachability theory and machine learning models to accurately forecast actor positions, leveraging sensors and map data to determine occupancy corridors and control vehicle movement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If data-driven prediction models are used to predict actor positions, then the model can output a finite number of hypotheses, but the model is not robust and does not cover the space of possible future positions well
Solution Approach 1:
The patent merges reachability theory-based prediction with data-driven prediction models. The reachability analysis provides a conservative bound on possible actor positions, while the data-driven model learns from historical data to predict likely positions. By combining these two approaches, the system achieves both robustness (from reachability) and adaptability (from data-driven learning), effectively covering the space of possible future positions while maintaining reliability.
Solution Approach 2:
The prediction system uses a composite approach by integrating two different prediction methodologies: reachability theory (which provides theoretical guarantees) and supervised learning models (which provide data-driven insights). This composite prediction framework leverages the strengths of both methods to produce predictions that are both reliable and adaptable to various driving scenarios.
2Reliability
If reachability-based actor prediction models are used, then the predictions are conservative, but they are overly-conservative and do not scale well in complex, chaotic, and congested driving environments
Solution Approach 1:
The patent applies partial reachability analysis by focusing on short-term predictions where reachability theory is most effective. For long-term predictions in complex environments, the system transitions to data-driven models that can scale better. This partial application of reachability theory maintains conservatism where needed while avoiding the scalability issues in complex, chaotic driving environments.
Solution Approach 2:
The system dynamically switches between reachability-based prediction and data-driven prediction based on the driving environment and time horizon. In simple, predictable scenarios, reachability theory provides conservative bounds. In complex, congested environments, the system leverages learned patterns from data to maintain scalability while still ensuring safety through the hybrid framework.
3Reliability
If reachability-based actor prediction models are used for long-term prediction, then the models do not scale well, but they provide conservative predictions
Solution Approach 1:
The patent segments the prediction time horizon into short-term and long-term components. For short-term predictions, reachability theory is used to provide conservative bounds on actor positions. For long-term predictions, data-driven models are employed to capture complex patterns and behaviors that emerge over extended periods. This segmentation allows the system to maintain conservatism where applicable while scaling to long-term prediction requirements.
Data Source
AI summary
A method for predicting an actor occupancy corridor includes receiving input data, predicting an occupancy sets of the actor using reachability analysis and the input data, determining an occupancy corridor constraints of the actor using a machine learning model and the input data, and determining an occupancy corridor of the actor using the occupancy corridor constraints of the actor using the machine learning model and the occupancy sets of the actor using a reachability analysis. Moreover, the method includes controlling the movement of a host vehicle based on the occupancy corridor of the actor.