Being good (at driving): Characterizing behavioral expectations on automated and human driven vehicles

(2) Santa Clara University; (3) West Virginia University; (4) University of Wisconsin-Madison); Francesca Favar\`o (1); Johan Engstrom (1); John Lee (4); Laura Fraade-Blanar (1); LLC; Melissa Cefkin (2); Ryan Best (3)

REVIEW 3 major objections 4 minor 2 cited by

Being good (at driving): Characterizing behavioral expectations on automated and human driven vehicles

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proposes Drivership, a framework that grounds good driving in the alignment between driving behavior and society's feasible normative expectations, applying to human and automated drivers alike.

desk verdict A well-written conceptual framework with a genuine new category, but the definition of Drivership contains an internal tension that needs to be resolved before the paper can serve as a benchmark for AV behavior. read the letter →

arxiv 2502.08121 v1 pith:GXIBKVZG submitted 2025-02-12 cs.CY cs.ET

Laura Fraade-Blanar (1) , Francesca Favarò (1) , Johan Engstrom (1) , Melissa Cefkin (2) , Ryan Best (3) , John Lee (4) , Trent Victor (1) ((1) Waymo , LLC

show 3 more authors

(2) Santa Clara University (3) West Virginia University (4) University of Wisconsin-Madison)

This is my paper · ORCID

classification cs.CYcs.ET

keywords Drivershipautomatedvehiclesdrivingbehaviorsocietalexpectationsnormativereferencemodelssafetycaseroadwaycitizenship

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces Drivership, a framework for saying what makes driving behavior good. Its central claim is that good driving is the realization of mutualistic expectations among road users: a driver performs well when its behavior aligns with what society believes drivers ought to do, within the limits of what is physically and technologically possible. To make this usable, the paper distinguishes Empirical Expectations, Normative Expectations, and a new category, Furtherance Expectations, and anchors Drivership in feasible societal Normative Expectations. Drivership covers both safety-centric behaviors and socially-aware behaviors with no direct safety stakes, and is meant to apply to humans and automated systems alike. If the framework holds, it offers a common vocabulary and benchmark basis for automated vehicle evaluation and for safety cases.

What carries the argument

The central object is a three-part taxonomy of expectations: Empirical Expectations, which are beliefs about what drivers will do based on past experience; Normative Expectations, which are beliefs about what drivers ought to do based on agreed-upon social principles; and Furtherance Expectations, which are beliefs about what could be done to continuously improve the transportation ecosystem. Drivership is positioned at the feasible subset of societal Normative Expectations, and behavior reference models, such as a quantitative model of a non-impaired driver with eyes on the conflict, are the machinery that turn those expectations into testable performance criteria.

What would settle it

Administer a representative survey of road users across several regions, presenting routine driving scenarios and asking what a driver ought to do; if no expectations achieve broad majority agreement and remain stable across repeated waves for core scenarios such as lane changes, yielding to pedestrians, and stopping distances, the framework's anchor in shared societal Normative Expectations would fail.

Watch

Extended reading notes

Core claim

On the paper's own terms, Drivership is the realization of good driving behavior, grounded not in a fixed list of virtues but in the alignment between driving behavior and societal expectations. The discovery is a way of organizing existing fragmented work: rather than asking which qualities define good driving, ask what road users owe each other and what society normatively expects. The paper argues that operationalization should be anchored in Normative Expectations that are shared at the societal level and feasible, and that these can be instantiated through behavior reference models, such as a model of a non-impaired driver's collision avoidance response. It also draws a sharp boundary for safety cases: behavior can be improper under Drivership without creating injury risk, and behavior with injury risk can still be proper Drivership, so safety evaluation must look at their intersection.

Load-bearing premise

The framework assumes that feasible societal Normative Expectations about how drivers ought to behave exist, are shared widely enough, and are stable enough to serve as evaluation benchmarks, but it offers no measurement procedure or evidence that such convergence occurs.

Editorial extensions

If this is right

Automated vehicle developers and regulators could benchmark behavior against societal Normative Expectations rather than against arbitrary attribute lists, making good driving measurable and comparable across systems.
Safety cases could separate improper Drivership from risk of harmful outcomes, concentrating certification evidence on events where both coincide.
The taxonomy gives value-sensitive design a shared vocabulary for surfacing conflicts among stakeholders' expectations, such as when Empirical and Normative expectations diverge.
Furtherance Expectations provide a mechanism for intentionally evolving norms, so automated vehicle behavior can be a deliberate force for improving roadway citizenship rather than only a responder to existing habits.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

An implication the authors leave implicit is that eliciting and tracking societal Normative Expectations should become a systematic measurement activity, using surveys, citizen panels, or behavioral studies, parallel to the collection of crash statistics.
The asymmetry that safe-but-uncourteous behavior can still qualify as Drivership, while courteous-but-unsafe behavior cannot, suggests that socially-aware failures will be treated as acceptance problems rather than safety defects, so automated vehicle deployments will likely prioritize courtesy only after safety baselines hold.
The same expectation-alignment logic could extend beyond the vehicle to infrastructure and policy, judging road designs and traffic laws by whether they match feasible societal Normative Expectations; the paper mentions infrastructure only in passing, leaving this extension open.
If norms shift with exposure to automated vehicles, Drivership benchmarks will need periodic revision, and an empirical study comparing stated driving expectations before and after prolonged automated vehicle operation in a city would test the framework's dynamic element.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper is a serious, well-structured conceptual contribution. What is actually new: it integrates Bicchieri's empirical/normative expectation distinction into a driving-specific framing ('Drivership') and adds a third category, Furtherance Expectations, which is a legitimate extension. The survey of behavioral reference models and the discussion of how Drivership could relate to safety cases are useful and mostly accurate. The writing is clear, the citations are appropriate, and the authors are honest about open questions, particularly around operationalization. Credit where due: this is a solid synthesis that could give the AV community a shared vocabulary.

The main soft spot is a real internal tension. The paper defines Drivership as alignment with societal Normative Expectations, then states that 'safe but uncourteous' behavior can attain Drivership while 'courteous but unsafe' cannot. If uncourteous behavior violates a Normative Expectation, then by the paper's own definition the driver is misaligned and has not fully attained Drivership. The safety-first priority rule is imported from outside the expectation-alignment framework, not derived from it. The paper acknowledges tensions between expectations but gives no resolution rule, and this specific statement is inconsistent with the central definition. This is fixable, but it needs to be addressed directly.

A second, milder concern is the operationalization gap: the paper assumes feasible societal Normative Expectations are identifiable and stable enough to serve as benchmarks, but offers no measurement procedure. This is acknowledged as future research, so it is more a limitation than a flaw.

The safety-case connection is argued rather than demonstrated, which is acceptable for a framing paper, but readers should not expect a concrete methodology.

Overall, the paper deserves serious review. It is a conceptual framework, not an empirical result, and its value depends on whether the community adopts the vocabulary. The internal inconsistency should be fixed before publication, but it does not sink the core idea.

I would bring this to a reading group focused on AV safety or value-sensitive design, and I would cite the expectation taxonomy if I were writing about AV behavioral benchmarking. Send it to peer review with a request for major revision on the Drivership definition and operationalization clarity.

Referee Report

3 major / 4 minor

Summary. The paper proposes a conceptual framework called Drivership, defined as the realization of good driving behavior, grounded in the alignment between a driver's behavior and the mutualistic expectations held among road users. Building on Bicchieri's distinction, it introduces Empirical Expectations and Normative Expectations, and adds a new category, Furtherance Expectations, to capture behaviors that could improve the transportation ecosystem. The paper positions Drivership within feasible societal Normative Expectations, distinguishes safety-centric from socially-aware behaviors, discusses operationalization through behavior reference models, and argues that Drivership supports the behavioral component of a safety case, specifically the determination of Absence of Unreasonable Risk. It also uses the 1899 Bliss crash as a motivating historical example and situates the framework in a broader theme of roadway citizenship.

Significance. If the framework is accepted, it offers a valuable coordinating concept for a fragmented literature on "good driving" in both human and automated driving contexts. Its strengths are the broad synthesis of sources across philosophy, standards, regulatory practice, and traffic safety research; the explicit driver-agnostic framing; the introduction of a vocabulary for expectations that can support value-sensitive design; and the conceptually useful linkage between driving behavior evaluation and safety-case reasoning. The paper is also commendably honest about several open questions, such as locality of expectations and the need for future research. However, the framework's central definition contains an internal tension about the priority of safety over other Normative Expectations, and the operationalizability of the key notion of "feasible societal Normative Expectations" is asserted rather than demonstrated. These issues bear directly on the paper's central claim and on its application to safety cases.

major comments (3)

[Section 3, paragraph beginning "In describing driving behaviors"] The statement that "it remains possible to attain Drivership through a safe but uncourteous behavior" is inconsistent with the paper's definition of Drivership as alignment with societal Normative Expectations. If courteous behavior is itself a societal Normative Expectation, then an uncourteous action is misaligned and should reduce, or preclude, Drivership. The paper gives no derivation of the safety-first priority from the expectation-alignment framework; it imports an external value ordering. This matters for Section 4, where behavioral Absence of Unreasonable Risk is tied to "valid societal moral concepts, denoted here as feasible Normative Expectations": if some Normative Expectations can be overridden by a safety criterion external to the expectation set, then the evaluation benchmark is not simply the mutually expected behavior. The authors should either redefine Drivership as alignment with a safety-compatible subset of Normative Expectations and justify that subset, or provide a resolution rule for conflicting expectations that is grounded in the framework itself.
[Sections 3.1 and 3.2] The framework assumes that feasible societal Normative Expectations are identifiable, sufficiently shared, and stable enough to serve as evaluation benchmarks, but no measurement or validation procedure is given. Footnote 36 cites criteria for "strong" and "stable" norms, and Section 3.2 mentions expert judgment and behavior reference models, yet there is no method for distinguishing a Normative Expectation from an Empirical one or for establishing that a given expectation holds at the societal level. This is load-bearing because Section 4 grounds behavioral AUR in feasible Normative Expectations. Without an elicitation or validation protocol, the claimed grounding of evaluation in mutualistic expectations is not operational. The paper should either describe such a procedure in sufficient detail, or explicitly reposition the contribution as a conceptual research program whose operational basis is an open question.
[Section 3.1, paragraph introducing Furtherance Expectations] The distinction between Furtherance Expectations and Normative Expectations is not operationally specified. The freeway-collision example is described as living in the overlap between Furtherance and Normative Expectations, but no criterion is given for when a proposed improvement is a Furtherance Expectation as opposed to an emerging Normative Expectation. Since the three-way taxonomy is a central contribution and is used to position Drivership in Figure 3, the paper should provide a demarcation rule or, failing that, explicitly argue why the boundary is intentionally fluid. As written, the new category risks being an unstipulated addition rather than a rigorously defined component of the framework.

minor comments (4)

[Footnote 2] The text contains a typo: "streetcar pick-usp" should read "pick-ups."
[Figure 2 caption] The caption says "societal norms and expectations," while the body of Section 3.1 carefully distinguishes Empirical, Normative, and Furtherance Expectations; the caption should use the paper's own taxonomy, e.g., "Normative Expectations," or explicitly refer to all three types.
[Section 3, safety-centric and socially-aware behaviors] The phrase "behaviors aligned with Drivership can have safety considerations but no social considerations, but not vice versa" is a clear statement, but it deserves a formal notation or example showing what counts as a social consideration without safety stakes; the current examples are vivid but the boundary is not systematically defined.
[Section 4] The phrase "proper (even if non-ideal) Drivership" would benefit from definitional clarity: if Drivership is a binary property of being realized, "proper Drivership" and "non-ideal Drivership" need to be defined in terms of the expectation-alignment framework, since the current wording suggests degrees that the main definition does not formalize.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the paper is an explicitly definitional framework, with its central claim stated as a proposal rather than derived from fitted data, self-citation chains, or renamed inputs.

full rationale

This is a conceptual/framework paper, not an empirical derivation. The central claim—that Drivership is the realization of good driving behavior grounded in alignment with feasible societal Normative Expectations—is introduced as an explicit definition and a stated choice ('We chose to anchor Drivership in Normative Expectations at the societal level, limited by what is feasible'), not as a prediction derived from data. No parameters are fitted, no closely related quantity is predicted from a fit, and no uniqueness theorem or prior result from the same authors is used to force the conclusion. The self-citations that appear (Roadmanship, NIEON, Waymo safety-process papers) function as prior inputs or illustrative operational examples, not as evidence validating the Drivership framework; the normative foundations are sourced to external authors (Bicchieri, Scanlon) and external standards (ISO, IEEE). The internal question raised by the safe-but-uncourteous passage in Section 3 is a coherence tension about whether safety trumps socially-aware normative expectations, but it is not a circularity: the paper does not use its own definitions as evidence for an empirical conclusion. The derivation chain is therefore self-contained with respect to circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 2 invented entities

The central claim rests on a normative postulate and on the validity of a social expectation taxonomy. No free parameters or empirical fits are involved. The framework's new categories are conceptual inventions, not entities with independent evidence.

assumptions (5)

ad hoc to paper Good driving behavior is properly defined by alignment with societal expectations rather than by outcomes or enumerated attributes.
Central postulate of Drivership, stated in Section 3: 'Drivership grounds the evaluation of driving behaviors in the alignment between the mutualistic expectations that exist amongst road users.'
domain assumption Bicchieri's distinction between Empirical and Normative Expectations is valid and sufficient as a base taxonomy.
Section 3.1 adopts these definitions from Bicchieri (2006) and Bicchieri et al. (2023).
ad hoc to paper A third category, Furtherance Expectations, is distinct from Normative Expectations and useful.
Introduced in Section 3.1 as a new contribution; the paper does not demonstrate that this category is non-redundant with Normative Expectations about behavior that 'could be exhibited to enable continuous improvement.'
domain assumption Scanlonian contractualism provides the correct ethical foundation for driving behavior.
Section 2 and Section 3.1 repeatedly invoke Scanlon (1998) and the idea of principles no one could reasonably reject.
domain assumption The laws of physics and current technological capabilities define the boundary of feasible expectations.
Section 3.1 defines Feasible Constraints with the stopping-distance example.

invented entities (2)

Drivership
purpose: Framing for the realization of good driving behavior as alignment between driving behavior and societal expectations.
New concept introduced; no falsifiable handle beyond its usability as a framing.
Furtherance Expectations
purpose: Third type of expectation regarding behaviors that could improve the transportation ecosystem.
New taxonomic category; no independent test distinguishes it from Normative Expectations.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Being good (at driving): Characterizing behavioral expectations on automated and human driven vehicles." pith.science (2026). https://pith.science/paper/GXIBKVZG

@misc{pith2026250208121,
  author       = {Pith},
  title        = {Pith review of: Being good (at driving): Characterizing behavioral expectations on automated and human driven vehicles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GXIBKVZG}},
  note         = {Machine review of arXiv:2502.08121}
}

read the original abstract

For over a century, researchers have wrestled with how to define good driving behavior, and the debate has surfaced anew for automated vehicles (AVs). We put forth the concept of Drivership as a framing for the realization of good driving behaviors. Drivership grounds the evaluation of driving behaviors in the alignment between the mutualistic expectations that exist amongst road users. Leveraging existing literature, we distinguish Empirical Expectations (i.e., reflecting "beliefs that a certain behavior will be followed," drawing on past experiences) (Bicchieri, 2006); and Normative Expectations (i.e., reflecting "beliefs that a certain behavior ought to be followed," based on societally agreed-upon principles) (Bicchieri, 2006). Because societal expectations naturally shift over time, we introduce a third type of expectation, Furtherance Expectations, denoting behavior which could be exhibited to enable continuous improvement of the transportation ecosystem. We position Drivership within the space of societal Normative Expectations, which may overlap with some Empirical and Furtherance Expectations, constrained by what is technologically and physically feasible. Additionally, we establish a novel vocabulary to rigorously tackle conversations on stakeholders' expectations, a key feature of value-sensitive design. We also detail how Drivership comprises safety-centric behaviors and what we term socially-aware behaviors (where there are no clear safety stakes). Drivership supports multiple purposes, including advancing the understanding and evaluation of driving performance through benchmarking based on many criteria. As such, we argue that an appropriate framing of the notion of Drivership also underpins the overall development of a safety case. The paper explores these applications under the more general tenet of Drivership as a central element to roadway citizenship.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Operationalization of Scenario-Based Safety Assessment of Automated Driving Systems
cs.RO 2025-07 conditional novelty 4.0 of 10

The authors propose an operationalized safety assessment framework that connects federated scenario databases to the NATM testing and approval process for automated vehicles.
What's Really Different with AI? -- A Behavior-based Perspective on System Safety for Automated Driving Systems
eess.SY 2025-07 conditional novelty 4.0 of 10

A position paper recommending that automated driving safety assurance separate AI-specific risks from open-context uncertainties and use behavior-based analyses to bridge them.

Reference graph

Works this paper leans on

6 extracted references · 3 canonical work pages · cited by 2 Pith papers

[1]

(2021, June 28)

ABC 30 Action News. (2021, June 28). Know the Road with the CHP: What to do when you have car problems or are in a crash on the freeway. ABC 30 Action News . https://abc30.com/driving-road-safety-chp-california-highway-patrol/10840461/ Alsaid, A., Lee, J. D., & Price, M. (2020). Moving Into the Loop: An Investigation of Drivers' Steering Behavior in Highl...

work page arXiv 2020
[4]

Concerning the Adoption of Harmonized Technical United Nations Regulations for Wheeled Vehicles, Equipment and Parts which can be Fitted and/or be Used on Wheeled Vehicles and the Conditions for Reciproc... UNECE. Retrieved September 5, 2024, from https://unece.org/transport/documents/2023/03/standards/un-regulation-no-157-amend4 Ecola, L., Popper, S. W.,...

arXiv 2018
[1182]

Acceptable Risk

https://local1182.org/about-us/history-of-traffic/ Oh, S., Avedisov, S., & Orosz, G. (2021). On the handling of automated vehicles: Modeling, bifurcation analysis, and experiments. European Journal of Mechanics , 90 (Nov-Dec). Oliveira, L., Proctor, K., Burns, C., & Birrell, S. (2019). Driving Style: How Should an Automated Vehicle Behave? Information , 1...

arXiv 2021
[1991]

Chaloux, N., & Ahmed, A

MMWR , 41 (24), 421-423. Chaloux, N., & Ahmed, A. (2019). Rules of the Road: Compliance and Defiance among the Different Types of Cyclists. Transportation Research Record , 2673 (9), 34-43. D'Amato, A., Dancel, S., Pilutti, J., Tellis, L., Frascaroli, E., & Gerdes, J. C. (2022). Exceptional Driving Principles for Autonomous Vehicles. Journal of Law and Mo...

arXiv 2019
[2018]

Rule-based Optimal Control for Autonomous Driving

The New York Times. Retrieved May 27, 2024, from https://www.nytimes.com/2018/03/30/nyregion/who-was-the-first-person-killed-by-a-car-i n-new-york.html Xiao, W., Mehdipour, N., Collin, A., Bin-Nun, A. Y., Frazzoli, E., Tebbens, R. D., & Belta, C. (2021). Rule-based Optimal Control for Autonomous Driving. arXiv . https://arxiv.org/abs/2101.05709 Yuen, K. F...

work page Pith review arXiv 2021
[4600]

ANSI/UL 4600 Standard for Safety for the Evaluation of Autonomous Products

(2023). ANSI/UL 4600 Standard for Safety for the Evaluation of Autonomous Products . UL Standards & Engagement. Retrieved October 16, 2024, from https://ulse.org/ul-standards-engagement/autonomous-vehicle-technology US DOT. (2011). The Haddon Matrix . Highway Safety Improvement Program Manual - Safety | Federal Highway Administration. Retrieved October 16...

work page arXiv 2023

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

(2021, June 28)

ABC 30 Action News. (2021, June 28). Know the Road with the CHP: What to do when you have car problems or are in a crash on the freeway. ABC 30 Action News . https://abc30.com/driving-road-safety-chp-california-highway-patrol/10840461/ Alsaid, A., Lee, J. D., & Price, M. (2020). Moving Into the Loop: An Investigation of Drivers' Steering Behavior in Highl...

work page arXiv 2020

[2] [4]

Concerning the Adoption of Harmonized Technical United Nations Regulations for Wheeled Vehicles, Equipment and Parts which can be Fitted and/or be Used on Wheeled Vehicles and the Conditions for Reciproc... UNECE. Retrieved September 5, 2024, from https://unece.org/transport/documents/2023/03/standards/un-regulation-no-157-amend4 Ecola, L., Popper, S. W.,...

arXiv 2018

[3] [1182]

Acceptable Risk

https://local1182.org/about-us/history-of-traffic/ Oh, S., Avedisov, S., & Orosz, G. (2021). On the handling of automated vehicles: Modeling, bifurcation analysis, and experiments. European Journal of Mechanics , 90 (Nov-Dec). Oliveira, L., Proctor, K., Burns, C., & Birrell, S. (2019). Driving Style: How Should an Automated Vehicle Behave? Information , 1...

arXiv 2021

[4] [1991]

Chaloux, N., & Ahmed, A

MMWR , 41 (24), 421-423. Chaloux, N., & Ahmed, A. (2019). Rules of the Road: Compliance and Defiance among the Different Types of Cyclists. Transportation Research Record , 2673 (9), 34-43. D'Amato, A., Dancel, S., Pilutti, J., Tellis, L., Frascaroli, E., & Gerdes, J. C. (2022). Exceptional Driving Principles for Autonomous Vehicles. Journal of Law and Mo...

arXiv 2019

[5] [2018]

Rule-based Optimal Control for Autonomous Driving

The New York Times. Retrieved May 27, 2024, from https://www.nytimes.com/2018/03/30/nyregion/who-was-the-first-person-killed-by-a-car-i n-new-york.html Xiao, W., Mehdipour, N., Collin, A., Bin-Nun, A. Y., Frazzoli, E., Tebbens, R. D., & Belta, C. (2021). Rule-based Optimal Control for Autonomous Driving. arXiv . https://arxiv.org/abs/2101.05709 Yuen, K. F...

work page Pith review arXiv 2021

[6] [4600]

ANSI/UL 4600 Standard for Safety for the Evaluation of Autonomous Products

(2023). ANSI/UL 4600 Standard for Safety for the Evaluation of Autonomous Products . UL Standards & Engagement. Retrieved October 16, 2024, from https://ulse.org/ul-standards-engagement/autonomous-vehicle-technology US DOT. (2011). The Haddon Matrix . Highway Safety Improvement Program Manual - Safety | Federal Highway Administration. Retrieved October 16...

work page arXiv 2023