[Paper Review] An "Xcity" Optimization Approach to Designing Proving Grounds for Connected and Autonomous Vehicles

Rui Chen, Mansur Arief+1 more|arXiv (Cornell University)|Aug 9, 2018

Autonomous Vehicle Technology and Safety12 references3 citations

TL;DR

This paper proposes an optimization framework called Xcity to design connected and autonomous vehicle (CAV) proving grounds by modeling traffic encounters as road assets and formulating the design as a nonlinear optimization problem. The approach maximizes scenario coverage and transition flexibility within spatial constraints, demonstrating effectiveness on Mcity road map data using BARON solver, though scalability is limited by computational complexity.

ABSTRACT

Proving ground, or on-track testing has been an essential part of testing and validation process for connected and autonomous vehicles (CAV). Several world-class CAV proving grounds, such as Mcity at the University of Michigan and The Castle of Waymo, have already been built, and many more are currently under construction. In this paper, we propose the first optimization approach to CAV proving ground designing and refer to any such CAV-centric design problem as "Xcity" to emphasize the enormous investment, the multi-dimensional spatial consideration, and the immense construction effort emerging globally. Inspired by the recent progress on traffic encounter clustering, we further define "road assets" as fundamental building blocks and formulate the whole design process into nonlinear optimization problems. We have shown that such framework can be utilized to adaptively generate CAV proving ground designs with optimized capability and flexibility and can further be extended to evaluate an existing "Xcity" design.

Motivation & Objective

• To address the lack of systematic design theory for CAV proving grounds despite massive global investment.
• To develop a method that maps real-world driving scenarios into space-constrained, controllable proving ground layouts.
• To optimize both testing capability (scenario coverage) and operational flexibility (transition between scenarios).
• To provide a scalable, mathematically grounded framework for evaluating and generating CAV testbed designs.
• To enable data-driven, scenario-based evaluation of CAV systems through infrastructure-level optimization.

Proposed method

• Models typical multi-vehicle CAV driving scenarios as 'road assets' derived from traffic encounter clustering and real-world data.
• Represents each road asset as a geometric and topological entity with nodes, segments, and connectivity constraints.
• Formulates a two-phase nonlinear optimization model: (1) feasible mapping of assets to space, (2) flexibility optimization via transition road placement.
• Uses bilinear and polynomial constraints to enforce orientation, connectivity, and intersection compatibility between assets.
• Employs a branch-and-bound solver (BARON) to solve the nonlinear mixed-integer program, with sensitivity analysis on solver performance.
• Applies the framework to Mcity’s road map, extracting assets and validating scalability on small-scale instances.

Experimental results

Research questions

• RQ1 How can typical CAV driving scenarios be systematically extracted and represented as modular road assets for proving ground design?
• RQ2 What optimization model can maximize both scenario coverage and transition flexibility in a space-constrained CAV proving ground layout?
• RQ3 How does the computational complexity of the optimization model scale with the number of road assets and internal segments?
• RQ4 Can the proposed framework be applied to evaluate or redesign an existing proving ground, such as Mcity?
• RQ5 Which nonlinear optimization solver performs best for this class of CAV proving ground design problems?

Key findings

• The Xcity framework successfully generates feasible and flexible CAV proving ground layouts using a scenario-based, optimization-driven approach.
• Feasible mapping for 1–3 road assets was solved in under 31 seconds on average, with increasing time as complexity grew.
• Flexibility optimization for up to 3 assets was computationally tractable, but problems with 4+ assets failed to find feasible solutions within 100 seconds.
• The model complexity grows significantly with the number of road assets and internal segments, dominated by bilinear and polynomial constraints.
• BARON outperformed other solvers (e.g., SCIP, Bonmin) in both feasibility and solution time, making it the preferred choice for this problem class.
• The framework demonstrates scalability limitations due to exponential growth in binary variables and nonlinear constraints, suggesting a need for decomposition methods in future work.