Autonomous Vehicle Steering Using Parallax-Based Obstacle Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous ground vehicles face challenges in navigating complex environments with obstacles, as existing methods often get stuck in local minima, fail to consider the vehicle's dimensions, and are computationally inefficient, especially when generating collision-free trajectories in real-time.
Innovation Solution
A method and apparatus that use parallax information from obstacle sensors to determine a second path avoiding obstacles, incorporating the vehicle's shape and dimensions, with a model predictive approach that minimizes a cost function including parallax costs and saturation constraints, allowing for real-time obstacle avoidance and efficient trajectory generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If reactive methods based on sensor data are used for local obstacle avoidance, then computational efficiency is improved, but the vehicle can get stuck in local minima and discretization of the world is required
Solution Approach 1:
The patent performs preliminary path planning to generate a reference path before real-time obstacle avoidance, and uses model predictive control to pre-calculate optimal trajectories that consider vehicle dynamics constraints. This preliminary action provides a foundation that reduces the complexity of real-time reactive decisions while maintaining reliability.
Solution Approach 2:
The patent introduces a model predictive control layer as an intermediary between simple reactive methods and complex global path planning. This intermediary uses a cost function that combines path tracking accuracy, vehicle dynamics constraints, and obstacle avoidance, producing smooth trajectories that avoid local minima while maintaining computational efficiency through efficient optimization algorithms.
2Device complexity
If the dimension of the vehicle is not explicitly considered in predictive active steering control, then the control algorithm is simpler, but in complex and cluttered environments, collision risk increases
Solution Approach 1:
The patent transitions from considering only the center of gravity position to explicitly incorporating vehicle dimensions by defining parallax regions based on the vehicle's bounding box. This dimensional expansion adds spatial awareness of the vehicle's physical extent, allowing the cost function to penalize trajectories that would cause collision with obstacles relative to the vehicle's actual dimensions rather than just its center point.
Solution Approach 2:
The patent applies different quality requirements to different parts of the vehicle by defining parallax regions for each vertex of the vehicle's bounding box. Each vertex has its own parallax cost calculation based on its position and the vehicle's orientation, allowing localized collision avoidance at critical points while maintaining overall trajectory smoothness.
3Speed
If optimization is performed within given sampling time of 10 ms or shorter for real implementation, then real-time control is achieved, but computational complexity must be reduced
Solution Approach 1:
The patent extracts and separates the computational tasks into different time scales: global path planning is performed offline or at lower frequency, while model predictive control handles real-time obstacle avoidance at 100 Hz. The cost function is designed to be computationally efficient by using closed-form solutions where possible and efficient numerical optimization for the remaining terms, enabling real-time performance.
Solution Approach 2:
The patent implements dynamic re-planning where the horizon length and optimization frequency adapt to the situation. In urgent situations with close obstacles, the controller performs more frequent optimizations with shorter horizons, while in safer conditions, it uses longer horizons with less frequent updates, optimizing the trade-off between responsiveness and computational load.
Data Source
AI summary
The present invention relates to an apparatus and a method for steering a vehicle, and more particularly to an apparatus and a method for steering an autonomous vehicle. One aspect of the present invention provides a method of steering a vehicle in an apparatus for steering the vehicle, the method comprising: (a) determining a first path by using a starting point and a destination of the vehicle; (b) generating an obstacle position signal when a plurality of obstacle sensors mounted on the vehicle detect an obstacle during an operation of the vehicle; (c) generating a parallax information using the obstacle position signal received from the plurality of the obstacle sensors; and (d) determining a second path avoiding the obstacle by using the parallax information. According to the method and the apparatus for steering a vehicle of the present invention, a vehicle may avoid obstacles in real-time in complex environments.