Vehicle Control Sequences for Adaptive Trajectory Planning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional vehicle control methods, such as those described in [reference 1], face limitations in efficiently determining control sequences that optimize trajectory planning and obstacle avoidance, particularly in dynamic environments, as they often rely on rigid sampling distributions that fail to adapt to changing conditions.

Innovation Solution

The method involves determining a first and second candidate control sequence, with the second sequence providing additional flexibility by using a weighted sum based on accumulated trajectory costs, and employing a conditional variational autoencoder (CVAE) to learn distributions that guide control decisions, allowing for more adaptive and efficient trajectory optimization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional rigid sampling distributions are used for control sequence determination, then computational simplicity is maintained, but adaptability to dynamic environments and trajectory optimization quality deteriorate

Engineering Contradiction:
Improveadaptability to dynamic environmentsVSAvoidcontrol method complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies dynamics by transitioning from rigid, static sampling distributions to dynamic, adaptive sampling distributions that evolve based on accumulated trajectory costs and environmental feedback. The sampling distribution is continuously updated during control sequence determination, allowing the system to adapt to changing conditions while maintaining computational tractability through structured update rules.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes parameters by modifying the sampling distribution parameters (mean and covariance) based on accumulated trajectory costs. Instead of using fixed distribution parameters, the system dynamically adjusts them to reflect learned task-aware characteristics, enabling better adaptation to dynamic environments while preserving the mathematical structure needed for efficient computation.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If multiple candidate control sequences are evaluated and combined using weighted sums, then trajectory optimization quality improves, but computational complexity increases

Engineering Contradiction:
Improvetrajectory optimization qualityVSAvoidcontrol sequence determination time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by evaluating multiple candidate control sequences with different levels of detail and combining them through weighted sums based on accumulated trajectory costs. Rather than exhaustively evaluating all possible sequences, the system strategically samples a subset of candidates and combines their contributions, achieving improved trajectory quality without proportional increases in computational time.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent implements feedback by using accumulated trajectory costs from previous evaluations to inform the selection and weighting of candidate control sequences. The system learns from past performance and uses this feedback to guide future control decisions, improving trajectory optimization quality over time while reducing redundant computations through experience-based prioritization.

Inventive Principle:
Principle #23Feedback

3Reliability

If task-aware distributions are learned and applied, then control solution quality improves, but system complexity and training requirements increase

Engineering Contradiction:
Improvecontrol solution qualityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by pre-training the system to learn task-aware distributions offline before actual control execution. The complex learning and adaptation work is performed in advance during a training phase, allowing the system to develop robust control policies that improve reliability. During runtime, the pre-learned distributions enable faster decision-making without the overhead of real-time learning complexity.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3739418B1Method of controlling a vehicle and apparatus for controlling a vehicle
Publication Date: 2021.12.15 ROBERT BOSCH GMBH
  • EP3739418B1 patent drawingFigure 1
  • EP3739418B1 patent drawingFigure 2~3
  • EP3739418B1 patent drawingFigure 4~7

AI summary

Method of controlling a vehicle or robot, wherein said method comprises the following steps: determining a first control sequence, determining a second control sequence for controlling said vehicle or robot depending on said first control sequence, a current state of said vehicle or robot, and on a model characterizing a dynamic behavior of said vehicle or robot, controlling said vehicle or robot depending on said second control sequence, wherein said determining of said first control sequence is performed depending on a first candidate control sequence and a second candidate control sequence.