Affiliation: University of Colorado Colorado Springs
Poster Title: Altruistic Routing Policies in Transportation Networks
Poster Abstract: It is well-known that selfish routing policies, where individual route time is minimized, from uncoordinated agents in a networked system can lead to inefficiencies in aggregate congestion. Likewise, it is known that if all traffic belongs to an autonomous fleet, and a system designer programs the fleet with altruistic routing policies---where routes are chosen in consideration of an individual's impact on aggregate congestion---optimal congestion emerges. However, if the system designer's fleet is only a fraction of the population, it is an open question as to when the presence of altruism improves aggregate congestion relative to homogeneous selfishness. To that end, we resolve this question definitively, and demonstrate that a series-parallel network, symmetric routes among populations, and a natural alignment between routes used by agents and the network (which we refer to as congruence), are necessary and sufficient conditions for a system designer to improve congestion by routing their fleet altruistically. When the network is series-parallel and congruent, but non-symmetric, we tightly bound the potential harm caused by altruistic routing in a heterogeneous network, relative to homogeneous selfishness. Further, when the system designer is able to modulate the altruism level of their fleet, we characterize the altruism level that balances reaching optimal efficiency with reducing the potential harm caused by heterogeneous routing.
Affiliation: University of Delaware
Poster Title: Distributed Optimization for Traffic Light Control and Connected Automated Vehicle Coordination in Mixed-Traffic Intersections
Poster Abstract: We consider the problem of coordinating traffic light systems and connected automated vehicles (CAVs) in mixed-traffic intersections. We aim to develop an optimization-based control framework that leverages both the coordination capabilities of CAVs at higher penetration rates and intelligent traffic management using traffic lights at lower penetration rates. Since the resulting optimization problem is a multi-agent mixed-integer quadratic program, we propose a penalization-enhanced maximum block improvement algorithm to solve the problem in a distributed manner. The proposed algorithm, under certain mild conditions, yields a feasible person-by-person optimal solution of the centralized problem. The performance of the control framework and the distributed algorithm is validated through simulations across various penetration rates and traffic volumes.
Affiliation: Johns Hopkins University
Poster Title: α-RACER: Real-Time Algorithm for Game-Theoretic Motion Planning and Control in Autonomous Racing
Poster Abstract: Autonomous racing extends beyond the challenge of controlling a racecar at its physical limits. Similar to human drivers, autonomous vehicles must also employ strategic maneuvers, such as overtaking and blocking, to gain an advantage over competitors. While modern control algorithms can achieve human-level performance in single-car scenarios, research on real-time algorithms for multi-car autonomous racing remains limited. To bridge this gap, we develop a game-theoretic modeling framework that incorporates the competitive aspects of autonomous racing, such as overtaking and blocking, through a novel policy parametrization, while operating the car at its limit. We propose an algorithmic approach to compute an approximate Nash equilibrium strategy for our game model, representing the optimal strategy for any autonomous vehicle in a competitive environment. Our approach leverages a recently introduced framework of dynamic α-potential functions, enabling the real-time computation. Our approach comprises two phases: offline and online. During the offline phase, we use simulated racing data to learn an α-potential function that approximates utility changes for agents. This function facilitates the online computation of approximate Nash equilibria by maximizing its value. We evaluate our method in a head-to-head 3-car racing scenario, demonstrating superior performance over several existing baselines. The paper is available online: https://arxiv.org/pdf/2412.08855.
Affiliation: National University of Singapore
Poster Title: Enforcing Cooperative Safety for Reinforcement Learning-based Mixed-Autonomy Platoon Control
Poster Abstract: It is recognized that the control of mixed-autonomy platoons comprising connected and automated vehicles (CAVs) and human-driven vehicles (HDVs) can enhance traffic flow. Among existing methods, Multi-Agent Reinforcement Learning (MARL) appears to be a promising control strategy because it can manage complex scenarios in real time. However, current research on MARL-based mixed-autonomy platoon control suffers from several limitations. First, existing MARL approaches address safety by penalizing safety violations in the reward function, thus lacking theoretical safety guarantees due to the black-box nature of RL. Second, few studies have explored the cooperative safety of multi-CAV platoons, where CAVs can be coordinated to further enhance the system-level safety involving the safety of both CAVs and HDVs. Third, existing work tends to make an unrealistic assumption that the behavior of HDVs and CAVs is publicly known and rationale. To bridge the research gaps, we propose a safe MARL framework for mixed-autonomy platoons. Specifically, this framework (i) characterizes cooperative safety by designing a cooperative Control Barrier Function (CBF), enabling CAVs to collaboratively improve the safety of the entire platoon, (ii) provides a safety guarantee to the MARL-based controller by integrating the CBF-based safety constraints into MARL through a differentiable quadratic programming (QP) layer, and (iii) incorporates a conformal prediction module that enables each CAV to estimate the unknown behaviors of the surrounding vehicles with uncertainty qualification. Simulation results show that our proposed control strategy can effectively enhance the system-level safety through CAV cooperation of a mixed-autonomy platoon with a minimal impact on control performance.
Affiliation: University of California Berkeley
Poster Title: Kernel-based Planning and Imitation Learning Control for Flow Smoothing in Mixed Autonomy Traffic
Poster Abstract: This work presents a new architecture for managing heterogeneous fleets aimed at achieving flow harmonization in mixed-autonomy traffic, demonstrating robustness across different sensing paradigms. We develop a kernel-based planning controller capable of providing anticipative coordination over low-bandwidth or high-latency networks. Furthermore, we employ a scenario based optimization technique to tune the parameters of the proposed controller which offers performance improvement over the grid search technique across different simulation scenarios. Additionally, our architecture includes a local control strategy utilizing imitation learning, distinctively treating our kernel-based planning controller as the expert. This unique application bridges the gap between local sensing and global sensing by introducing input flexibility and vehicle control decentralization while preserving behavioral alignment with the expert’s potential actions. Our proposed architecture is shown to be adaptable across a broad spectrum of car platforms, accommodating vehicles with varying levels of sensing and actuation, highlighting its potential for widespread implementation in future transportation systems.
Affiliation: ETH Zurich
Poster Title: Co-investment with Payoff Sharing Benefit Operators and Users in Network Design
Poster Abstract: Network-based complex systems are inherently interconnected, with the design and performance of subnetworks being interdependent. However, the decisions of self-interested operators may lead to suboptimal outcomes for users. In this paper, we consider the question of what cooperative mechanisms can benefit both operators and users simultaneously. We address this question in a game theoretical setting, integrating both non-cooperative and cooperative game theory. During the non-cooperative stage, subnetwork decision-makers strategically design their local networks. In the cooperative stage, the co-investment mechanism and the payoff-sharing mechanism are developed to enlarge collective benefits and fairly distribute them. A case study of the Sioux Falls network is conducted to demonstrate the efficiency of the proposed framework. The impact of this interactive network design on environmental sustainability, social welfare, and economic efficiency is evaluated, along with an examination of scenarios involving regions with heterogeneous characteristics.
Affiliation: University of Washington
Poster Title: Learning Responsibility Allocations for Multi-agent Interactions: A Differentiable Optimization Approach with Control Barrier Functions
Poster Abstract: From autonomous driving to package delivery, ensuring safe yet efficient multi-agent interaction is challenging as the interaction dynamics are influenced by hard-to-model factors such as social norms and contextual cues. Understanding these influences can aid in the design and evaluation of socially-aware autonomous agents whose behaviors are aligned with human values. In this work, we seek to codify factors governing safe multi-agent interactions via the lens of responsibility, i.e., an agent's willingness to deviate from their desired control to accommodate safe interaction with others. Specifically, we propose a data-driven modeling approach based on control barrier functions and differentiable optimization that efficiently learns agents' responsibility allocation from data. We demonstrate on synthetic and real-world datasets that we can obtain an interpretable and quantitative understanding of how much agents adjust their behavior to ensure the safety of others given their current environment.