Backward-Looking Surprise Metric for Autonomous Agent Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Fully autonomous driving by autonomous vehicles is challenging due to the difficulty in predicting the actions of other agents, such as pedestrians and cyclists, which are not directly observable and affect decision-making.

Innovation Solution

A system computes a backward-looking surprise metric by comparing vehicle data at a current time step to predicted trajectories from a previous time step, using a generative model to assess the surprise of an agent's behavior, both for the vehicle itself and others, enabling online and offline analysis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the system uses a backward-looking surprise metric comparing posterior state to prior probability distribution, then prediction accuracy and realism are improved, but computational complexity increases

Engineering Contradiction:

Improveprediction accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system pre-computes prior probability distributions of agent behaviors before actual driving decisions are needed. By preparing these probability models in advance based on historical data and simulation, the system reduces real-time computational burden while maintaining high prediction accuracy when evaluating surprise metrics during actual operation.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system uses simulated agent behaviors and virtual driving scenarios to create representative copies of real-world driving patterns. These simulated priors capture the essence of agent behavior distributions without requiring complex real-time analysis of actual agent intentions, thereby simplifying computation while preserving prediction accuracy.

Inventive Principle:

Principle #26Copying

2Reliability

If the system computes surprise metrics for both current and future states using multiple predicted trajectories, then decision-making quality is improved, but processing time increases

Engineering Contradiction:

Improvedecision-making qualityVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system computes surprise metrics for a selected subset of predicted trajectories rather than exhaustively evaluating all possible trajectories. By focusing computation on the most probable or critical trajectories identified through preliminary filtering, the system achieves sufficient decision-making quality without the prohibitive processing time required for complete trajectory enumeration.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12576889B2Method, system, and media for calculating a surprise score by comparing a posterior state to a prior probability distribution

Publication Date: 2026.03.17 WAYMO LLC

AI summary

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for computing a backward looking surprise metric for autonomously driven vehicles. One of the methods includes obtaining first data representing one or more previously predicted states of an agent along one or more predicted trajectories of the agent at a first time step. Second data representing one or more states of the agent at a subsequent time step is obtained. A surprise score is computed from a measure of a difference between the first data computed for the one or more predicted trajectories for the prior time step and the second data computed for the one or more predicted states for the subsequent time step.