Tractor-Trailer Path Planning With Neural Off-Tracking Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current path/trajectory planners for autonomous tractor-trailer combinations fail to consider the large dimensions and articulated structures of these vehicles, leading to infeasible trailer paths and off-tracking issues, especially during sharp turns. Additionally, existing complex planners require significant computational resources to meet real-time requirements.
Innovation Solution
A deep neural network with an encoder-decoder architecture is trained to generate feasible paths for tractor-trailer combinations, minimizing off-tracking and trailer swept areas while avoiding collisions with static obstacles. The training utilizes a path cost function that evaluates the 'true' cost of a path objectively, eliminating the need for expert driving data and enabling training on fake data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If current path/trajectory planners for normal autonomous cars are applied to autonomous semi-trucks, then the planning process is simple, but the trailer paths become infeasible and go off track
Solution Approach 1:
The patent applies local quality by training the neural network to generate different paths for different parts of the vehicle system. The network learns to account for the articulated structure by generating tractor paths that consider trailer constraints, ensuring each component (tractor and trailer) has appropriate path characteristics for its specific requirements while maintaining overall system feasibility
Solution Approach 2:
The patent changes parameters by incorporating trailer-specific parameters into the path planning process. The neural network is trained to consider trailer dimensions, articulation angles, and swept area constraints as key parameters, transforming the planning from simple car-based geometry to complex tractor-trailer kinematics that ensure feasible trailer paths
2Reliability
If complex MPC-based path planning algorithms are used for autonomous semi-trucks, then the trailer path feasibility is improved, but the computational resources required increase significantly
Solution Approach 1:
The patent replaces the mechanical/computational system of MPC-based optimization algorithms with a neural network-based system. Instead of using iterative mathematical optimization that requires significant computational power, the solution uses a trained neural network that directly generates feasible paths through pattern recognition, substituting heavy computation with learned behavioral patterns
Solution Approach 2:
The patent uses copying by training the neural network on simulated data that replicates real-world driving scenarios. The network learns from copied experiences in simulation environments, allowing it to generalize to real-world situations without requiring expensive real-time computational optimization during actual operation
3Measurement precision
If expert driving data is collected for training path planners, then the training accuracy is improved, but the data collection process becomes complex and time-consuming
Solution Approach 1:
The patent applies self-service by using a path cost function that automatically evaluates and guides the training process without requiring expert driving data. The cost function provides objective feedback during training, allowing the system to self-correct and learn optimal paths through automated evaluation rather than manual expert demonstrations
Solution Approach 2:
The patent achieves universality by creating a training system that works without specialized expert data. The path cost function serves multiple purposes: evaluating path quality, guiding network training, and ensuring feasibility constraints are met, making the training process universally applicable without requiring specific expert-driven datasets
Data Source
AI summary
A path planning approach based on semi-supervised learning including a trained encoder-decoder type of deep neural network to generate and plan paths with the objective to minimize the off-track of the tractor-trailer swept area. The encoder encodes input information such as lane markings, static obstacles, and potentially other features, and pass it to the decoder to generate a planned path. A path cost function scores and penalizes each network-generated path based on its deviation from the lane center, the path smoothness and collision with any static obstacles, and backpropagates the cost of the paths through the encoder-decoder network to train it. As the path cost function acts as a critic of the path quality, no collected data from expert driving for training is required, but only randomly generated samples of many possible combinations of lane shapes and obstacles arrangements.