Autonomous Vehicle Trajectory Forecasting Across Multi-Actor Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in reliably interpreting and responding to their surrounding environment, particularly in complex scenarios involving multiple actors and dynamic navigation paths.
Innovation Solution
A method for training machine learning models that identifies actors and streams in the vehicle's environment, generates trajectories for the vehicle and actors, and updates these trajectories iteratively using a stream connection function to improve decision-making and control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If autonomous vehicles use basic sensor data processing and simple navigation algorithms, then the system complexity is low and ease of operation is high, but the reliability of interpreting complex environmental scenarios and making accurate navigation decisions deteriorates
Solution Approach 1:
The patent segments the complex navigation problem into distinct components: identifying individual actors (pedestrians, vehicles, cyclists) as separate entities, determining their respective streams (candidate navigation paths) independently, and then processing trajectory forecasts for each actor-stream combination separately. This segmentation allows the system to handle complex environments by breaking them down into manageable units that can be processed reliably.
Solution Approach 2:
The patent introduces a new dimensional framework by conceptualizing navigation in terms of multiple streams (candidate paths) rather than single trajectories. Each actor is associated with multiple possible streams, and the system evaluates interactions across these dimensional layers. This dimensional expansion allows the system to consider multiple potential futures simultaneously, improving reliability in complex scenarios.
2Measurement precision
If autonomous vehicles forecast trajectories for multiple actors and streams without iterative updates, then the processing time and computational resources are reduced, but the accuracy of trajectory forecasting and decision-making deteriorates
Solution Approach 1:
The patent performs preliminary trajectory forecasting for each actor with respect to their associated stream before iterative updates begin. This initial forecast establishes a baseline that guides subsequent iterative refinements. By having a preliminary action in place, the system ensures that even if iterative processing is truncated, a reasonable trajectory estimate exists, while still allowing for accuracy improvements when full iteration is feasible.
Solution Approach 2:
The patent implements iterative updates where trajectory forecasts are continuously refined based on feedback from evaluating actor-stream interactions. Each iteration uses information from previous iterations to improve accuracy, creating a feedback loop that progressively enhances measurement precision. The system balances this feedback-driven accuracy improvement against computational time constraints by adapting the number of iterations based on scenario complexity.
3Reliability
If autonomous vehicles consider all possible actor-stream interactions, then the comprehensiveness of environmental interpretation is improved, but the device complexity and computational burden increase
Solution Approach 1:
The patent creates a universal processing framework that handles all actor-stream interactions through a single standardized methodology. The same stream identification, trajectory forecasting, and interaction evaluation processes are applied universally across all actors and streams, regardless of their specific types or roles. This universal approach maintains comprehensiveness while managing complexity through consistency and reusability of processing modules.
Solution Approach 2:
The patent introduces streams as intermediary constructs that mediate between raw sensor data about actors and the final navigation decisions. Rather than directly processing all pairwise actor interactions, the system uses streams as intermediate representations that organize and structure the information. These intermediary streams simplify the complexity of multi-actor interactions by providing a structured framework for evaluation.
Data Source
AI summary
Implementations process, using machine learning (ML) layer(s) of ML model(s), actor(s) from a past episode of locomotion of a vehicle and stream(s) in an environment of the vehicle during the past episode to forecast associated trajectories, for the vehicle and for each of the actor(s), with respect to a respective associated stream of the stream(s). Further, implementations process, using a stream connection function, the associated trajectories to forecast a plurality of associated trajectories, for the vehicle and each of the actor(s), with respect to each of the stream(s). Moreover, implementations iterate between using the ML layer(s) and the stream connection function to update the associated trajectories for the vehicle and each of the actor(s). Implementations subsequently use the ML layer(s) in controlling an AV.