Autonomous Vehicle Path Planning Using Inverse Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles often generate trajectories that lack smoothness, resulting in abrupt maneuvers that are less desirable for passengers due to static weights in conventional path planning systems.
Innovation Solution
An open space path planning method using Inverse Reinforcement Learning (IRL) to dynamically adjust weights based on environmental conditions, mimicking human driving behavior by applying a reward function derived from expert driver behaviors, thereby optimizing the path planning system to produce smoother trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional path planning with static weights is used, then the path planning system is simple and computationally efficient, but the generated trajectories lack smoothness and produce abrupt maneuvers
Solution Approach 1:
The patent applies dynamics by transitioning from static weights to dynamic weights that adapt based on environmental conditions. The IRL model continuously learns optimal weights from expert driver behaviors and adjusts them in real-time according to the current driving context, enabling the path planning system to generate smoother trajectories while responding to changing environmental conditions.
Solution Approach 2:
The patent implements parameter changes by modifying the weight parameters in the cost function based on environmental conditions. Instead of using fixed weights, the system dynamically adjusts weight values to prioritize different cost components (e.g., safety, comfort, efficiency) depending on the current situation, thereby improving trajectory smoothness and human-like driving behavior.
2Ease of operation
If static weights are used in the cost function, then the path planning system is computationally efficient, but the maneuvers performed are abrupt and less desirable for passengers
Solution Approach 1:
The patent applies preliminary action by pre-training the IRL model offline using expert driver behaviors to learn optimal weight configurations for various environmental conditions. This pre-computed knowledge is stored and then quickly retrieved and applied during real-time path planning, avoiding the need for time-consuming online learning while still achieving smooth, comfortable trajectories.
3Adaptability or versatility
If dynamic weight adjustment based on environmental conditions is implemented, then the trajectories become smoother and more human-like, but the path planning system complexity increases
Solution Approach 1:
The patent applies copying by using the IRL model to copy and replicate expert driver behaviors. The system observes and learns from demonstrated expert driving trajectories, extracting the underlying decision-making patterns and weight configurations that produce smooth, human-like maneuvers, and then applies these copied behaviors to autonomous path planning.
Data Source
AI summary
In one embodiment, a method determines a route from a first location of an autonomous driving vehicle (ADV) to a second location within an open space, the first location being a current location of the ADV. The method determines an objective function based on the route, the objective function having a set of costs for maneuvering the ADV from the first location to the second location. The method determines environmental conditions of the open space and uses the environmental conditions to determine a set of weights, each weight to be applied to a corresponding cost of the objective function. The method optimizes the objective function in view of one or more constraints, such that an output of the objective function reaches minimum while the one or more constraints are satisfied and generates a path trajectory with the optimized objective function to control the ADV autonomously according to the path trajectory.