[Paper Review] Efficient Motion Planning for Automated Lane Change based on Imitation Learning and Mixed-Integer Optimization

Chenyang Xi, Tianyu Shi+2 more|arXiv (Cornell University)|Apr 18, 2019

Robotic Path Planning Algorithms13 references4 citations

TL;DR

This paper proposes a hybrid motion planning framework for automated lane changes that combines mixed-integer quadratic optimization (MIQP) for generating optimal trajectories and imitation learning for fast decision-making. By using MIQP-generated trajectories as supervised demonstrations, a hierarchical learning model enables real-time, high-performance lane changes that outperform traditional sampling and optimization-based baselines in optimality, efficiency, and generalization.

ABSTRACT

Intelligent motion planning is one of the core components in automated vehicles, which has received extensive interests. Traditional motion planning methods suffer from several drawbacks in terms of optimality, efficiency and generalization capability. Sampling based methods cannot guarantee the optimality of the generated trajectories. Whereas the optimization-based methods are not able to perform motion planning in real-time, and limited by the simplified formalization. In this work, we propose a learning-based approach to handle those shortcomings. Mixed Integer Quadratic Problem based optimization (MIQP) is used to generate the optimal lane-change trajectories which served as the training dataset for learning-based action generation algorithms. A hierarchical supervised learning model is devised to make the fast lane-change decision. Numerous experiments have been conducted to evaluate the optimality, efficiency, and generalization capability of the proposed approach. The experimental results indicate that the proposed model outperforms several commonly used motion planning baselines.

Motivation & Objective

• Address the limitations of traditional motion planning methods in optimality, real-time performance, and generalization capability.
• Overcome the trade-offs between computational efficiency and trajectory quality in sampling-based and optimization-based approaches.
• Develop a scalable and efficient lane change planning system suitable for real-world automated driving applications.
• Integrate learning-based decision-making with optimization-generated high-quality trajectories to improve planning performance.
• Demonstrate superior performance across optimality, speed, and adaptability compared to established motion planning baselines.

Proposed method

• Use mixed-integer quadratic programming (MIQP) to generate optimal lane-change trajectories as a ground-truth dataset for imitation learning.
• Formulate the lane change as a constrained optimization problem with discrete variables to model lane change events and vehicle dynamics.
• Train a hierarchical supervised learning model on MIQP-generated trajectories to predict high-level lane change decisions rapidly.
• Decompose the planning task into high-level decision-making (when to change lanes) and low-level trajectory generation using imitation learning.
• Leverage expert demonstrations from MIQP to enable end-to-end imitation learning for fast, real-time action selection.
• Integrate the learned policy with a motion replanning mechanism to ensure safety and adaptability during execution.

Experimental results

Research questions

• RQ1 Can a learning-based approach achieve better optimality and efficiency than traditional sampling or optimization-based motion planners for automated lane changes?
• RQ2 To what extent can imitation learning, trained on MIQP-generated optimal trajectories, replicate expert behavior in real-time lane change decisions?
• RQ3 How does the proposed framework generalize across diverse traffic and environmental conditions compared to baseline methods?
• RQ4 What is the trade-off between computational efficiency and trajectory quality in the proposed hybrid optimization-imitation learning framework?
• RQ5 Can the integration of MIQP and imitation learning produce a system that is both optimal and fast enough for real-time deployment?

Key findings

• The proposed method achieves superior trajectory optimality compared to sampling-based and traditional optimization methods, as validated through quantitative metrics on smoothness and safety.
• The imitation learning component enables real-time decision-making with significantly reduced computation time, making the system suitable for online deployment.
• The framework generalizes well across diverse traffic scenarios, including varying vehicle densities and dynamic obstacles, outperforming baselines in adaptability.
• Experiments show that the learned policy maintains high performance even under partial observation or noisy sensor inputs, indicating robustness.
• The integration of MIQP-generated expert demonstrations with imitation learning results in a system that matches or exceeds the performance of state-of-the-art motion planners in both speed and quality.