Vehicle Trajectory Planning Using Scenario Trees and Interaction Intentions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for planning vehicle driving trajectories in autonomous driving are inadequate for strong and long-tail interaction scenarios, as they fail to accurately predict the intentions of other traffic agents, leading to unreliable and unsafe driving trajectories.

Innovation Solution

A method that simulates interaction scenarios of a vehicle at current and future moments using a scenario tree, determines optimal interaction intentions based on a reinforcement learning model trained on human driving behavior, and uses these intentions to plan safe and reliable driving trajectories.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If preset rules are used to predict intentions of other traffic agents, then the method works for common scenarios, but it fails to accurately predict intentions in strong interaction scenarios and long-tail interaction scenarios

Engineering Contradiction:
Improvereliability of driving trajectoryVSAvoidadaptability to interaction scenarios
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system pre-trains a reinforcement learning model using extensive human driving behavior data before deployment. This preliminary training enables the model to learn and adapt to various interaction scenarios including strong interaction and long-tail scenarios, allowing accurate intention prediction when actual driving situations arise without requiring preset rules for each scenario

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system transitions from using fixed preset rules to using a dynamic reinforcement learning model that can adapt its predictions based on learned patterns from training data. The model's parameters are optimized through training on human driving behaviors, enabling it to handle diverse scenarios including those with strong interactions and long-tail cases where preset rules fail

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If preset rules are used for each long-tail interaction scenario, then accurate intentions can be obtained, but the complexity of presetting rules for each scenario becomes unmanageable

Engineering Contradiction:
Improveaccuracy of intention predictionVSAvoidcomplexity of rule system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Instead of creating complex preset rules for each long-tail scenario, the system uses a reinforcement learning model trained on copied human driving behavior data. The model learns from patterns in the training data and generalizes to new scenarios, avoiding the need to explicitly program rules for each specific scenario while maintaining accurate intention prediction

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The reinforcement learning model serves as a universal solution that handles multiple types of interaction scenarios including common scenarios, strong interaction scenarios, and long-tail scenarios with a single system. This multi-functional approach eliminates the need for separate preset rules for each scenario type, significantly reducing system complexity

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4527705B1Method for determining vehicle driving trajectory, computer device, storage medium, and vehicle
Publication Date: 2025.12.03 ANHUI NIO AUTONOMOUS DRIVING TECH CO LTD
  • EP4527705B1 patent drawingFigure 1
  • EP4527705B1 patent drawingFigure 2
  • EP4527705B1 patent drawingFigure 3~4

AI summary

The disclosure relates to the technical field of autonomous driving, and provides a method for determining a vehicle driving trajectory, a computer device, a storage medium, and a vehicle, to solve the problem of improving the reliability and safety in planning vehicle driving trajectories. The method includes: sequentially simulating, based on an interaction scenario of a vehicle at a current moment by using forward driving of the vehicle as a constraint condition, an interaction scenario of the vehicle at each of a plurality of future moments, to form a scenario tree consisting of the interaction scenarios at the current moment and the future moments; obtaining a scenario value of each interaction intention in the interaction scenario at each moment in the scenario tree; obtaining an optimal interaction intention of the vehicle in the interaction scenario at each moment based on the scenario value; and determining a driving trajectory of the vehicle based on the optimal interaction intention. According to the disclosure, the optimal interaction intention of the vehicle can be made more consistent with correct driving habits of humans, thereby ensuring the autonomous driving safety and reliability of the vehicle.