Agent Trajectory Ranking for Low-Latency Self-Driving Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-driving vehicles face challenges in accurately predicting the motion of agents like pedestrians or human-driven vehicles, due to high variance in possible future behaviors and uncertainties in intent and realization.
Innovation Solution
The technology uses a pre-built trajectory bank containing feature vectors associated with observed agent trajectories. Sensor data from the self-driving vehicle is used to generate a current situation feature vector, which is then searched against the trajectory bank to identify relevant trajectories for predicting the agent's motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a motion prediction system generates predictions from scratch to capture high variance in agent behavior, then prediction accuracy improves, but computational complexity and latency increase
Solution Approach 1:
The system pre-computes and stores a trajectory bank containing feature vectors and associated trajectories offline before real-time operation. During runtime, the system only needs to perform similarity matching between current situation feature vectors and stored feature vectors, rather than generating predictions from scratch. This preliminary action significantly reduces computational complexity and latency while maintaining prediction accuracy.
2Reliability
If a motion prediction system computes multiple possible trajectories to quantify uncertainty, then reliability of prediction improves, but processing time increases
Solution Approach 1:
The trajectory bank is pre-computed offline to include multiple possible trajectories with their associated feature vectors. During real-time operation, the system quickly retrieves relevant trajectories from the pre-computed bank based on similarity matching, avoiding the need to compute multiple trajectories from scratch during time-critical processing.
Solution Approach 2:
Instead of generating new trajectory predictions during runtime, the system copies relevant trajectories from the pre-computed trajectory bank that match the current situation. This copying approach maintains reliability by providing multiple possible trajectories while minimizing processing time.
3Measurement precision
If the system searches through all possible trajectories to find the most likely ones, then prediction accuracy improves, but computational efficiency deteriorates
Solution Approach 1:
The system transforms the trajectory prediction problem into a feature vector similarity matching problem. By representing situations and trajectories as feature vectors and using distance metrics to measure similarity, the system can efficiently search through the trajectory bank using optimized algorithms rather than evaluating all possible trajectories, thus maintaining accuracy while improving computational efficiency.
Data Source
AI summary
A method of operating a self-driving vehicle is provided. The method includes predicting a trajectory of an agent in a vicinity of the self-driving vehicle. This is done by: receiving sensor data indicative of a current situation in the vicinity of the self-driving vehicle; generating a current situation feature vector based, at least in part, on the sensor data; searching a pre-built bank of library feature vectors to identify a result feature vector from the pre-built bank that is most relevant to the current situation feature vector, wherein each library feature vector in the pre-built bank is associated with an observed trajectory; and predicting the trajectory of the agent based, at least in part, on the result feature vector. The method also includes planning an action of the vehicle based, at least in part, on the predicted agent trajectory, and operating the vehicle according to the planned action.