Robust Learning-Based Trajectory Planning for Emerging Mobility Systems

Behdad Chalaki, Andreas A. Malikopoulos

This paper, titled "Robust Learning-Based Trajectory Planning for Emerging Mobility Systems," presents a significant step forward in making autonomous intersections safe for the real world. By combining Gaussian Process (GP) regression with robust optimization, the authors provide a way for Connected and Automated Vehicles (CAVs) to coordinate safely even when they can't follow their planned paths perfectly.

TL;DR

The authors develop a decentralized coordination framework that uses Gaussian Process regression to learn and predict vehicle trajectory errors in real-time. By incorporating these learned uncertainty bounds into a robust optimization problem, they guarantee collision-free intersection crossings even in the presence of sensor noise and tracking errors.

What Problem Does This Paper Address?

Coordinating CAVs at signal-free intersections promises to eliminate stop-and-go traffic, saving fuel and time. Most existing algorithms assume "perfect" execution—that a vehicle will be exactly where it planned to be at every millisecond.

In reality, low-level controller errors, sensor noise, and environmental factors cause "nominal" trajectories to drift. If two vehicles are scheduled to pass through a conflict point with a narrow safety margin, even a small deviation can lead to a collision. This paper addresses the gap between idealized deterministic planning and the messy, uncertain reality of physical vehicle dynamics.

Methods

The authors extend a previously developed "single-level" coordination framework. The core of their new approach is a three-step process:

  1. Learning Uncertainty (GP Regression): As a CAV moves toward an intersection, it collects noisy observations of its actual position vs. its planned position. It uses Gaussian Process (GP) regression with a Matern 3/2 kernel to model the error function.
  2. Propagating Errors: The paper provides mathematical proofs to map the error in "time-at-position" to errors in speed and position. Because speed deviations follow a unimodal Chi-square distribution, the authors use the Vysochanskii-Petunin inequality to create tighter confidence intervals than standard methods (like Chebyshev’s inequality) would allow.
  3. Robust Reformulation: The original optimization problem (minimizing exit time) is rewritten. Instead of satisfying safety constraints for a single point, the vehicle must satisfy them for the entire "tube" of possible trajectories defined by the 95% confidence interval.

The system is decentralized: each vehicle characterizes its own uncertainty and shares its "robust" trajectory with a central coordinator, which acts as a shared database for other incoming vehicles.

Datasets

The paper relies on numerical simulations rather than a pre-existing public dataset.

  • Scenario: A signal-free intersection with 24 CAVs entering from 6 different paths.
  • Parameters: Traffic flow of 3600 vehicles/hour; initial speeds between 12–14 m/s.
  • Uncertainty Model: The authors simulated actual deviations using a logarithmic function ($0.012 \log(1 + p_i)^{1.5}$) plus Gaussian noise to test the GP's ability to learn unknown error patterns.

Key Insights and Ideas

  • The "Safety Tube" Concept: Instead of planning a line, the algorithm plans a "tube." If the tubes of two vehicles don't collide, the physical vehicles are statistically guaranteed (at a 95% confidence level) to remain safe.
  • Distribution-Aware Bounds: A major "aha moment" is the realization that speed deviations aren't just random; they are unimodal. By applying the Vysochanskii-Petunin inequality, the authors achieve safer coordination without being overly conservative (which would happen if they used a one-size-fits-all inequality).
  • Online Replanning: The framework includes a "replanning event." When a vehicle reaches a specific "uncertainty characterization point," it updates its model and broadcasts the new, more accurate trajectory to all following vehicles, who then adjust their own plans.

Key Results

  • Safety Guarantee: In simulations, the robust framework successfully prevented lateral and rear-end collisions.
  • Failure of Deterministic Models: The authors showed that under the same uncertainty conditions, standard deterministic coordination (ignoring error) resulted in vehicles crossing the "lateral constraint" lines—essentially causing simulated crashes.
  • Efficiency: Despite the added complexity of robust constraints, the system still allows for high-throughput, signal-free coordination, proving that "robustness" doesn't have to mean "slow."

Limitations and Future Work

  • Computational Complexity: While the authors aim for real-time performance, solving non-linear robust optimization problems for every vehicle can be taxing. The paper notes that iterative solvers can be "computationally expensive."
  • Communication Assumptions: The model assumes perfect communication between CAVs and the coordinator. In the real world, latency and packet loss could introduce a new layer of uncertainty not covered here.
  • Validation: The results are currently limited to numerical simulations. The authors suggest that validating this on an experimental testbed (physical scale-model CAVs) is a necessary next step.

Summary updated

Abstract

In this paper, we extend a framework that we developed earlier for coordination of connected and automated vehicles (CAVs) at a signal-free intersection to incorporate uncertainty. Using the possibly noisy observations of actual time trajectories and leveraging Gaussian process regression, we learn the bounded confidence intervals for deviations from the nominal trajectories of CAVs online. Incorporating these confidence intervals, we reformulate the trajectory planning as a robust coordination problem, the solution of which guarantees that constraints in the system are satisfied in the presence of bounded deviations from the nominal trajectories. We demonstrate the effectiveness of our extended framework through a numerical simulation.
BibTeX
@inproceedings{chalaki2022robust,
  author = {Behdad Chalaki and Andreas A. Malikopoulos},
  title = {{Robust Learning-Based Trajectory Planning for Emerging Mobility Systems}},
  booktitle = {2022},
  year = {2022},
  doi = {10.23919/acc53348.2022.9867265},
}
Figure