Vehicle Trajectory Biasing for Human-Aware Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotic planning systems struggle to accurately predict vehicle actions in dynamic environments, particularly in scenarios involving objects and human-driven behaviors, leading to potential safety and efficiency issues.
Innovation Solution
A computing device employs multiple machine learned models, one trained on cost-based data and another on human driving data, to generate and select a trajectory that balances safety and efficiency by determining a bias value based on sensor and map data, optimizing computational resources and leveraging human driving behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single machine learned model is used to predict vehicle actions, then the system complexity is low, but the prediction accuracy and safety in dynamic environments deteriorate
Solution Approach 1:
The patent combines multiple machine learned models (cost-based model and human-driven behavior model) into a unified planning system. The cost-based model generates trajectories optimized for safety and efficiency, while the human-driven behavior model predicts actions of other road users. By merging these complementary models, the system achieves higher prediction accuracy and reliability compared to using a single model alone.
Solution Approach 2:
The planning system is designed to perform multiple functions simultaneously: it generates candidate trajectories, evaluates costs, predicts human behaviors, and selects optimal actions. This multi-functional approach allows a single system to handle diverse driving scenarios (lane changes, intersections, pedestrian interactions) with high accuracy without requiring separate specialized systems for each function.
2Reliability
If multiple machine learned models are employed to generate planned trajectories, then the prediction accuracy improves, but the computational resource consumption increases
Solution Approach 1:
The patent segments the trajectory generation and selection process into distinct stages handled by different models. The cost-based model segments trajectory generation into candidate trajectory evaluation, while the human-driven behavior model segments prediction into human action forecasting. This segmentation allows computational resources to be distributed efficiently across specialized sub-tasks rather than using a single computationally intensive model for all functions.
Solution Approach 2:
The system generates multiple candidate trajectories and evaluates them partially through cost analysis and human behavior prediction, rather than exhaustively evaluating all possible trajectories. The bias value mechanism allows the system to selectively focus computational resources on promising trajectories, performing excessive action only where needed to achieve sufficient prediction accuracy without unnecessary computational overhead.
3Adaptability or versatility
If the system uses human driving data to train a model, then the adaptability to human behaviors improves, but the training data requirements and system complexity increase
Solution Approach 1:
The patent applies different training data characteristics to different models based on their specific functions. The human-driven behavior model is trained specifically on human driving data to capture local patterns of human behavior, while the cost-based model uses simulation data and cost functions optimized for safety-critical decisions. This local quality approach allows each model to be highly adaptable to its specific task without requiring all models to process all types of training data.
Solution Approach 2:
The planning system acts as an intermediary that integrates predictions from the human-driven behavior model with cost-based trajectory evaluation. Rather than directly controlling the vehicle based solely on human behavior predictions, the system mediates between human behavior predictions, safety constraints, and efficiency objectives through the cost function and bias value mechanism, achieving adaptability while maintaining safety guarantees.
Data Source
AI summary
Techniques for determining a planned trajectory usable to control a vehicle in an environment are discussed herein. A computing device can receive multiple planned trajectories generated by different models, and determine to use one of the planned trajectories to control the vehicle at a future time. The models can represent machine learned models that are independently trained using different training data and one of the models may leverage human driving data during training. The techniques can also include determining a bias value to cause the vehicle to utilize a planned trajectory from a set of available planned trajectories.