pith. sign in

arxiv: 2605.23782 · v1 · pith:D3KJUS2Vnew · submitted 2026-05-22 · 💻 cs.GT · cs.SY· eess.SY

Routing Equilibrium in Mixed-Autonomy Traffic Networks with Altruistic Autonomous Agents

Lihui Yi , Ermin Wei This is my paper

Pith reviewed 2026-05-25 02:20 UTC · model grok-4.3

classification 💻 cs.GT cs.SYeess.SY
keywords mixed-autonomy trafficrouting gamevariational inequalityaltruistic autonomous agentssocial costequilibrium uniquenesstraffic assignment
0
0 comments X

The pith

Formulating the mixed routing game as a variational inequality proves equilibrium existence without convexity and uniqueness of aggregate flows for certain costs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper studies a traffic assignment problem where human drivers minimize their own travel times while autonomous vehicles minimize total social cost. By casting the resulting equilibrium as a variational inequality, the authors establish existence without requiring convex cost functions and obtain uniqueness of the aggregated link flow together with the social cost when costs belong to a specific class. The same variational inequality then supplies sufficient conditions that determine whether adding autonomous vehicles will reduce, raise, or leave unchanged the overall social cost. The framework also shows that a centralized planner directing the autonomous vehicles reaches the identical equilibrium when costs are convex.

Core claim

The equilibrium of the simultaneous routing game between self-interested human drivers and altruistic autonomous agents is characterized as the solution to a variational inequality. This formulation establishes existence of equilibrium without assuming convex costs and guarantees uniqueness of the aggregated link flow and social cost under a specific class of cost functions. The framework also supplies sufficient conditions under which adding autonomous agents improves, deteriorates, or has no effect on social cost, and shows equivalence to centralized routing under convex costs.

What carries the argument

The variational inequality formulation of the mixed routing game equilibrium.

If this is right

  • Equilibrium exists even when cost functions are not convex.
  • Aggregated link flows and social cost are unique at equilibrium under the identified class of cost functions.
  • Explicit sufficient conditions separate the cases in which autonomous agents improve social cost, worsen it, or leave it unchanged.
  • Centralized routing of autonomous agents produces the same equilibrium as decentralized choice when costs are convex.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the altruism model holds in practice, traffic authorities could program fleets of autonomous vehicles to target the regimes where social cost decreases.
  • The uniqueness of aggregate flows may allow simpler computation or control of network-level outcomes than full game-theoretic equilibria.
  • The conditions could be tested numerically on real road networks by varying the autonomous fraction and observing social-cost trends.

Load-bearing premise

Autonomous agents are programmed to minimize total social cost rather than their own travel times.

What would settle it

A network example in which the aggregated link flow is not unique at equilibrium even though the cost functions satisfy the specific class used for uniqueness, or an instance where the social cost moves in the opposite direction from the predicted sufficient conditions when the fraction of autonomous agents increases.

Figures

Figures reproduced from arXiv: 2605.23782 by Ermin Wei, Lihui Yi.

Figure 3
Figure 3. Figure 3: Social cost vs. fraction of autonomous agents (deterioration case, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Social cost and aggregated path flow vs. fraction of autonomous [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
read the original abstract

Recent advancements in vehicle autonomy have drawn interest in understanding the impact of autonomous vehicles on traffic systems. In this paper, we study a traffic assignment problem in a mixed-autonomy setting where both human-driven and autonomous vehicles coexist. We model the interaction as a simultaneous routing game where human drivers are self-interested and aim to minimize their own travel times, while autonomous agents are altruistic and aim to minimize the total social cost. The standard nonatomic congestion game analysis establishes the existence of equilibrium to this game under convex cost functions, and does not have any implication of its uniqueness. In this work, we formulate the equilibrium as a variational inequality (VI), which enables us to establish the equilibrium existence without convexity assumption, and guarantees the uniqueness of the aggregated link flow and social cost at equilibrium under a specific class of cost functions. Leveraging this VI framework, we provide sufficient conditions under which including autonomous agents improves, deteriorates, or has no effect on social cost. While the possibility of deterioration has been established in prior work, our results complement existing worst-case bounds by explicitly characterizing sufficient conditions under which each outcome occurs, thereby providing a deeper understanding of mixed-autonomy traffic systems. Furthermore, we consider a centralized scenario where a social planner optimizes the routing of autonomous agents, and show that the same equilibrium is achieved as in the decentralized scenario when assuming convex costs.Finally, we conduct numerical experiments that illustrate how social cost changes with the amount of autonomous vehicles under different system parameters.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper models a mixed-autonomy nonatomic routing game in which human drivers minimize individual latency while altruistic autonomous vehicles minimize total social cost. It reformulates the resulting equilibrium as a variational inequality (VI) to prove existence without requiring convexity of the latency functions, establishes uniqueness of the aggregate link flow and social cost under a specific (unspecified in the abstract) class of cost functions, derives sufficient conditions under which the introduction of AVs improves, deteriorates, or leaves unchanged the social cost, shows that the decentralized equilibrium coincides with the social planner's solution when costs are convex, and illustrates the comparative-static results numerically.

Significance. If the claimed VI equivalence holds and the uniqueness class is non-vacuous, the work supplies an analytically tractable framework that moves beyond worst-case bounds by giving explicit sufficient conditions for the sign of the social-cost effect; the centralized-decentralized equivalence under convexity is also a useful sanity check. These contributions would be of interest to the traffic-assignment and algorithmic game-theory communities.

major comments (3)
  1. [Abstract / game definition] Abstract and game-definition paragraph: the claim that the mixed Nash equilibrium is equivalent to a single standard VI is load-bearing for all subsequent results, yet the precise mapping is not exhibited. The human players satisfy the standard VI <c(f), g_H − f_H> ≥ 0 while the AVs satisfy <∇SC(f), g_AV − f_AV> ≥ 0; these coincide with one VI only if ∇SC is a simple function of the latency vector c. Without an explicit statement of the combined operator and verification that it inherits the continuity/coercivity needed for existence, the assertion of “existence without convexity” cannot be assessed.
  2. [Abstract] Abstract: the “specific class of cost functions” guaranteeing uniqueness of aggregate flow and social cost is never named, nor is any verification supplied that the class is non-vacuous or that the uniqueness proof does not tacitly re-introduce convexity. Because uniqueness is used to obtain the comparative-static conditions, this omission renders the central claims unverifiable from the given description.
  3. [Abstract] Abstract: the sufficient conditions for improvement/deterioration of social cost are stated to be derived from the VI framework, but without the explicit VI operator or the cost-class definition it is impossible to check whether those conditions are independent of the modeling premise that AVs are programmed to minimize SC.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful and constructive report. We address each major comment below and will revise the manuscript to improve clarity on the variational inequality formulation, the cost-function class, and the derivation of the comparative-static conditions.

read point-by-point responses

  1. Referee: [Abstract / game definition] Abstract and game-definition paragraph: the claim that the mixed Nash equilibrium is equivalent to a single standard VI is load-bearing for all subsequent results, yet the precise mapping is not exhibited. The human players satisfy the standard VI <c(f), g_H − f_H> ≥ 0 while the AVs satisfy <∇SC(f), g_AV − f_AV> ≥ 0; these coincide with one VI only if ∇SC is a simple function of the latency vector c. Without an explicit statement of the combined operator and verification that it inherits the continuity/coercivity needed for existence, the assertion of “existence without convexity” cannot be assessed.

    Authors: We agree that the abstract and introductory game-definition paragraph would benefit from greater explicitness. Section 3 of the manuscript defines the equilibrium as the single VI with operator F(f) whose human components are the latency vector c(f) and whose AV components are ∇SC(f). We will add a concise statement of this combined operator to the abstract and game-definition paragraph, together with a short verification that F is continuous (by continuity of c and SC) and coercive on the feasible set (by standard growth assumptions on latencies), which holds without requiring convexity of the latency functions. revision: yes

  2. Referee: [Abstract] Abstract: the “specific class of cost functions” guaranteeing uniqueness of aggregate flow and social cost is never named, nor is any verification supplied that the class is non-vacuous or that the uniqueness proof does not tacitly re-introduce convexity. Because uniqueness is used to obtain the comparative-static conditions, this omission renders the central claims unverifiable from the given description.

    Authors: The class in question is the set of latency functions that are strictly monotone in the aggregate link flow (Definition 4.1). We will name the class explicitly in the abstract and add a brief paragraph verifying that the class is non-vacuous (linear latencies belong to it) and that the uniqueness argument relies only on strict monotonicity of the aggregate operator, without invoking convexity. revision: yes

  3. Referee: [Abstract] Abstract: the sufficient conditions for improvement/deterioration of social cost are stated to be derived from the VI framework, but without the explicit VI operator or the cost-class definition it is impossible to check whether those conditions are independent of the modeling premise that AVs are programmed to minimize SC.

    Authors: The comparative-static conditions are obtained by comparing the VI at different AV penetration rates under the monotonicity properties of the combined operator; they therefore inherit the modeling premise that AVs minimize SC. We will revise the abstract to state the operator and cost class explicitly, thereby making the derivation traceable and allowing readers to verify the logical dependence. revision: yes

Circularity Check

0 steps flagged

No circularity: standard game-to-VI reformulation with explicit modeling assumptions

full rationale

The paper states the altruism modeling premise explicitly as an assumption rather than deriving it. It then applies the standard variational inequality reformulation to the resulting mixed routing game. No parameters are fitted from data and relabeled as predictions, no self-citation chains support load-bearing steps, and uniqueness is conditioned on an explicitly stated class of cost functions. The derivation chain is self-contained against external benchmarks and does not reduce any claimed result to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The model rests on standard nonatomic routing-game assumptions plus the domain choice that AVs are altruistic; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • standard math Existence of Nash equilibrium in nonatomic congestion games under standard continuity assumptions
    Invoked to establish baseline existence before the VI reformulation.
  • domain assumption Autonomous agents are programmed to minimize aggregate social cost rather than individual travel time
    Central modeling choice stated in the game definition; enables the VI and comparative-static results.

pith-pipeline@v0.9.0 · 5798 in / 1344 out tokens · 35354 ms · 2026-05-25T02:20:05.772148+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [1]

    Effects of autonomous vehicle ownership on trip, mode, and route choice,

    M. W. Levin and S. D. Boyles, “Effects of autonomous vehicle ownership on trip, mode, and route choice,”Transportation Research Record, vol. 2493, no. 1, pp. 29–38, 2015

    work page 2015

  2. [2]

    Preparing for the Future of Transportation: Automated Vehicles 3.0,

    U.S. Department of Transportation, “Preparing for the Future of Transportation: Automated Vehicles 3.0,” https://www.transportation. gov/av/3, 2018, accessed: Dec. 25, 2025

    work page 2018

  3. [3]

    Stabilizing traffic flow via a single autonomous vehicle: Possibilities and limitations,

    S. Cui, B. Seibold, R. Stern, and D. B. Work, “Stabilizing traffic flow via a single autonomous vehicle: Possibilities and limitations,” in2017 IEEE Intelligent Vehicles Symposium (IV), 2017, pp. 1336–1341

    work page 2017

  4. [4]

    Emergent behaviors in mixed-autonomy traffic,

    C. Wu, A. Kreidieh, E. Vinitsky, and A. M. Bayen, “Emergent behaviors in mixed-autonomy traffic,” inProceedings of the 1st Annual Conference on Robot Learning, vol. 78. PMLR, 13–15 Nov 2017, pp. 398–407

    work page 2017

  5. [5]

    Social behavior for autonomous vehicles,

    W. Schwarting, A. Pierson, J. Alonso-Mora, S. Karaman, and D. Rus, “Social behavior for autonomous vehicles,”Proceedings of the Na- tional Academy of Sciences, vol. 116, no. 50, pp. 24 972–24 978, 2019

    work page 2019

  6. [6]

    Safety-critical traffic control for mixed autonomy systems with input delay and disturbances,

    C. Zhao and H. Yu, “Safety-critical traffic control for mixed autonomy systems with input delay and disturbances,” in2023 62nd IEEE Conference on Decision and Control (CDC), 2023, pp. 1094–1099

    work page 2023

  7. [7]

    Social coordination and altruism in autonomous driving,

    B. Toghi, R. Valiente, D. Sadigh, R. Pedarsani, and Y . P. Fallah, “Social coordination and altruism in autonomous driving,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 12, pp. 24 791– 24 804, 2022

    work page 2022

  8. [8]

    A game theoretic macro- scopic model of lane choices at traffic diverges with applications to mixed–autonomy networks,

    N. Mehr, R. Li, and R. Horowitz, “A game theoretic macro- scopic model of lane choices at traffic diverges with applications to mixed–autonomy networks,”Transportation Research Part B: Method- ological, vol. 144, pp. 45–59, 2021

    work page 2021

  9. [9]

    Regional cooperative decision-making based on coalition game for multilane merging in mixed traffic,

    M. Fu, S. Li, M. Guo, X. Wang, X. Li, and W. Wang, “Regional cooperative decision-making based on coalition game for multilane merging in mixed traffic,”IEEE Transactions on Intelligent Trans- portation Systems, vol. 26, no. 12, pp. 22 665–22 679, 2025

    work page 2025

  10. [10]

    Coupled vehicle-signal control based on stackelberg game enabled multi-agent reinforcement learning in mixed traffic environment,

    X. Zhang, Z. He, Y . Zhu, and W. Huang, “Coupled vehicle-signal control based on stackelberg game enabled multi-agent reinforcement learning in mixed traffic environment,”Physica A: Statistical Mechan- ics and its Applications, vol. 658, p. 130289, 2025

    work page 2025

  11. [11]

    Two-sided deep reinforcement learning for dynamic mobility-on-demand management with mixed autonomy,

    J. Xie, Y . Liu, and N. Chen, “Two-sided deep reinforcement learning for dynamic mobility-on-demand management with mixed autonomy,” Transportation Science, vol. 57, no. 4, pp. 1019–1046, 2023

    work page 2023

  12. [12]

    Some theoretical aspects of road traffic research,

    J. G. Wardrop, “Some theoretical aspects of road traffic research,” Proceedings of the Institution of Civil Engineers, vol. 1, no. 3, pp. 325–362, May 1952

    work page 1952

  13. [13]

    Multiclass traffic assignment model for mixed traffic flow of human-driven vehicles and connected and au- tonomous vehicles,

    J. Wang, S. Peeta, and X. He, “Multiclass traffic assignment model for mixed traffic flow of human-driven vehicles and connected and au- tonomous vehicles,”Transportation Research Part B: Methodological, vol. 126, pp. 139–168, 2019

    work page 2019

  14. [14]

    Reliability-based mixed traffic equilib- rium problem under endogenous market penetration of connected autonomous vehicles and uncertainty in supply,

    Q. Zhong and L. Miao, “Reliability-based mixed traffic equilib- rium problem under endogenous market penetration of connected autonomous vehicles and uncertainty in supply,”Networks and Spatial Economics, vol. 24, pp. 461–505, 03 2024

    work page 2024

  15. [15]

    Equilibrium modeling of mixed autonomy traffic flow based on game theory,

    J. Li, D. Chen, and M. Zhang, “Equilibrium modeling of mixed autonomy traffic flow based on game theory,”Transportation Research Part B: Methodological, vol. 166, pp. 110–127, 2022

    work page 2022

  16. [16]

    The price of anarchy for transportation networks with mixed autonomy,

    D. A. Lazar, S. Coogan, and R. Pedarsani, “The price of anarchy for transportation networks with mixed autonomy,” in2018 Annual American Control Conference (ACC), 2018, pp. 6359–6365

    work page 2018

  17. [17]

    Altruistic auton- omy: Beating congestion on shared roads,

    E. Bıyık, D. A. Lazar, R. Pedarsani, and D. Sadigh, “Altruistic auton- omy: Beating congestion on shared roads,” inAlgorithmic Foundations of Robotics XIII. Springer International Publishing, 2020, pp. 887– 904

    work page 2020

  18. [18]

    Can taxes improve congestion on all networks?

    P. N. Brown and J. R. Marden, “Can taxes improve congestion on all networks?”IEEE Transactions on Control of Network Systems, vol. 7, no. 4, pp. 1643–1653, 2020

    work page 2020

  19. [19]

    The tradeoff between altruism and anarchy in transportation networks,

    C. Hill and P. N. Brown, “The tradeoff between altruism and anarchy in transportation networks,” in2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC), 2023, pp. 1442–1447

    work page 2023

  20. [20]

    Traffic equilibrium and variational inequalities,

    S. Dafermos, “Traffic equilibrium and variational inequalities,”Trans- portation Science, vol. 14, no. 1, pp. 42–54, 1980

    work page 1980

  21. [21]

    Bureau of Public Roads,Traffic Assignment Manual for Application with a Large, High Speed Computer. U.S. Department of Commerce, Urban Planning Division, 1964

    work page 1964

  22. [22]

    M. Hadi, X. Zhou, and D. Hale,Multiresolution Modeling for Traffic Analysis: Case Studies Report. U.S. Department of Transportation, Federal Highway Administration, 2022

    work page 2022

  23. [23]

    Flow count data-driven static traffic assignment models through network modularity partition- ing,

    A. Roocroft, G. Punzo, and M. A. Ramli, “Flow count data-driven static traffic assignment models through network modularity partition- ing,”Transportation, vol. 52, 09 2023

    work page 2023

  24. [24]

    Mixed transportation network design under a sustainable development perspective,

    J. Qin, L.-L. Ni, and F. Shi, “Mixed transportation network design under a sustainable development perspective,”The Scientific World Journal, vol. 2013, no. 1, p. 549735, 2013

    work page 2013

  25. [25]

    Nagurney,Network Economics

    A. Nagurney,Network Economics. Springer Science & Business Media, Mar. 2013

    work page 2013

  26. [26]

    A traffic assignment model based on link densities,

    L. de Grange, M. Marechal, and F. Gonz ´alez, “A traffic assignment model based on link densities,”Journal of Advanced Transportation, no. 1, 2019

    work page 2019

  27. [27]

    Beckmann, C

    M. Beckmann, C. B. McGuire, and C. B. Winsten,Studies in the Economics of Transportation. Yale University Press, 1956

    work page 1956

  28. [28]

    The traffic assignment problem for multiclass-user transportation networks,

    S. Dafermos, “The traffic assignment problem for multiclass-user transportation networks,”Transportation Science, vol. 6, no. 1, pp. 73–87, 1972

    work page 1972

  29. [29]

    On a paradox of traffic planning,

    D. Braess, A. Nagurney, and T. Wakolbinger, “On a paradox of traffic planning,”Transportation Science, vol. 39, no. 4, pp. 446–450, 2005. APPENDIXI PROOF OFTHEOREM1 Here, we prove the general case with multiple O/D pairs. The proof for the single O/D pair case follows directly. For this purpose, we need to first refine our notations. LetJ denote the set o...

    work page 2005

  30. [30]

    The maximum number of iterations is bounded by the total number of paths

    The algorithm will terminate in finite time, since at each iteration, either one path is removed fromP A (step 10) or at least one path is added toP H (step 12). The maximum number of iterations is bounded by the total number of paths

    work page

  31. [31]

    Paths inP H always have the same travel time cost, which is also the minimum among all paths (step 12)

    work page

  32. [32]

    Only paths inP H have non-zero flow at the end of the algorithm, due to the termination condition (step

    work page

  33. [33]

    and the update rule (step 8)

    work page

  34. [34]

    Therefore, the output path flow pattern˜xis an equilibrium for the baseline system

    During the entire process, no path will be removed fromP H, and path flows on paths inP H only increase (step 8). Therefore, the output path flow pattern˜xis an equilibrium for the baseline system. Moreover, we have˜xp ≥x H∗ p for all p∈ P. Thus, by Lemma 2, we haveS ∗ ≤ ˜S∗. The inequality is strict as long asx ∗ ̸= ˜x∗. The rest of the proof is to show ...

    work page

  35. [35]

    Autonomous agents use pathq= arg min p∈P C A p (˜x∗)

    work page

  36. [36]

    Specifically, for autonomous agents, we letx A =αe q, where eq is the unit vector with1at theq-th entry and0elsewhere

    Human agents use and only use paths inV, while main- taining the travel time cost on these paths equalized. Specifically, for autonomous agents, we letx A =αe q, where eq is the unit vector with1at theq-th entry and0elsewhere. For human agents, we have the following equation: ∆T V K∆(x H +x A) +b =λ H 1. Since human agents only use paths inV, we havex H p...

    work page

  37. [37]

    Human agents have no incentive to use paths outside V, i.e.,C H p (xH , xA)≥λ H for allp /∈ V

    work page

  38. [38]

    Human agents flow pattern is feasible, i.e.,x H ≥0

    work page

  39. [39]

    Autonomous agents have no incentive to use paths other thanq, i.e.,C A p (xH , xA)≥C A q (xH , xA)for all p∈ P. Condition (1): For any pathp /∈ V, we have C H p (xH , xA) = ∆T (Kf+b) p = ∆T K∆(x H +x A) + ∆T b p = ∆T K∆ V xH V + ∆T K∆e qα+ ∆ T b p = ∆T K∆ V(˜x∗ V +αd) + ∆ T K∆e qα+ ∆ T b p =C H p (˜x∗) +α ∆T K∆ V d+ ∆ T K∆e q p. (16) Sinceλ H = ˜λH +αγ, w...

    work page