A Systematic Study of Multi-Agent Deep Reinforcement Learning for Safe and Robust Autonomous Highway Ramp Entry

Larry Schester; Luis E. Ortiz

REVIEW 4 major objections 4 minor 42 references

A Systematic Study of Multi-Agent Deep Reinforcement Learning for Safe and Robust Autonomous Highway Ramp Entry

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Multi-agent reinforcement learning controllers trained by self-play achieve near-ideal collision performance for highway ramp merging with more than two vehicles, according to this simulation study.

desk verdict A legitimate incremental extension of the authors' two-vehicle MA-DRL merging work to three-vehicle and full-scene settings, but the 'nearly ideal' claim rests on an undocumented ideal baseline and post hoc network selection, so it is weaker than advertised. read the letter →

arxiv 2411.14593 v2 pith:G55WXWR6 submitted 2024-11-21 cs.RO cs.AIcs.LGcs.MAcs.SYeess.SY

Larry Schester , Luis E. Ortiz This is my paper · ORCID

classification cs.ROcs.AIcs.LGcs.MAcs.SYeess.SY

keywords multi-agentreinforcementlearningdeepautonomousdrivinghighwayrampmergingself-playcollisionavoidancelongitudinalcontrolgametheory

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

This paper argues that deep reinforcement learning controllers trained through multi-agent self-play can make an autonomous vehicle merge onto a highway almost as safely as is theoretically possible, even when several other vehicles are present. Prior work proved that in a fully decentralized two-vehicle merge, two rational agents cannot guarantee zero collisions. The authors extend the setting to three-vehicle scenes and to a fuller scene with two merging vehicles and a stream of traffic, and they measure the learned controllers' collision frequencies against an idealized 'best possible' collision table. Their simulation results show near-ideal collision rates for typical gap and ramp-length settings, suggesting that this style of game-theoretic training is a viable path for automated ramp merging.

What carries the argument

The mechanism is the combination of a game-theoretic self-play training loop and a reductionist scene model. The ego merge vehicle and the reactive traffic vehicles are all represented by actor-critic DDPG networks; rewards strongly penalize collisions and reward successful merges. State variables include closing gaps and closing speeds to the nearest front and rear vehicles, time-to-goal, and a time-in-between-vehicles measure for traffic. The full-scene model avoids exponential scaling by always identifying and adapting to the closest and most relevant vehicles, which lets a fixed-size state vector serve arbitrarily large scenes. The benchmark is an idealized collision table showing which ramp-length and starting-differential cells are unavoidable even under the most favorable action selection; the learned controllers' collision percentages are then compared cell by cell to this table.

What would settle it

Recompute the idealized collision table under the same protocol used to test the learned controllers, specifically the same constant, random, and reactive traffic policies, the same TIV settings, and the same initial conditions, and compare the learned controllers' collision percentages cell by cell against this recomputed table. If the learned controllers exceed the recomputed baseline in any cell where the original table predicted zero avoidable collisions, the 'nearly ideal' claim is weakened.

Watch

Extended reading notes

Core claim

The central claim is that controllers learned by deep deterministic policy gradient in a self-play multi-agent setting are nearly ideal when measured against an idealized optimal controller for highway ramp merging with more than two vehicles. The paper constructs an intermediate three-vehicle scene (one merging vehicle, two traffic vehicles) and a full scene (two merging vehicles, two or more traffic vehicles), using a reductionist model that tracks only the closest relevant vehicles to avoid exponential growth in state and action spaces. In standardized tests with constant, random, and reactive traffic policies, the learned policies achieve collision rates close to the 'best possible' table of Fig. 7 for gap sizes of 15 m and above, and they avoid collisions well within typical ramp lengths. The authors take this as empirical support that near-optimal performance carries over from the previously established two-vehicle setting to more realistic multi-vehicle scenes.

Load-bearing premise

The comparison to the idealized 'best possible' table assumes that this table is a valid upper bound for the exact test protocol used with the learned controllers, but the paper does not describe how the table is computed nor whether those idealized conditions match the same non-cooperative traffic policies used during testing.

Editorial extensions

If this is right

If the claim holds, multi-agent self-play RL can produce merging controllers whose collision rates in simulation approach the theoretical floor, without requiring explicit vehicle-to-vehicle coordination.
The reductionist, closest-vehicle modeling approach suggests a fixed-size state representation that can be applied to highways with arbitrarily many vehicles, since only the nearest front and rear vehicles matter for longitudinal control.
The results indicate that longitudinal-only control (acceleration and deceleration) may be sufficient for safe merging at typical ramp lengths, deferring the need for steering and lane-changing in this function.
The strong reward shaping with massive penalties for collisions may be a general recipe for safety-critical driving tasks where the cost of failure dominates.

Reading between the lines

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

A hidden risk is that the 'ideal' table might assume cooperative or omniscient behavior, whereas the test traffic uses myopic constant, random, or reactive policies; if so, some cells in the learned controllers' table could be worse than the true achievable optimum under the same test protocol.
The approach could be extended to mixed traffic with human drivers by treating human controllers as one of the traffic policies, provided a suitable human behavior model is available.
A testable prediction is that adding steering and lane-changing to the state set will not change the near-ideal longitudinal performance at gap sizes above 15 m, because those gaps are large enough that the second traffic vehicle's influence is already negligible.

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

Quick take: this is a legitimate incremental extension of Schester and Ortiz's earlier two-vehicle work, scaling to three-vehicle and full-scene merge scenarios using multi-agent DDPG with a sensible 'reductionist' state selection. The empirical study is systematic and the authors are candid about training instability and model selection challenges. However, the central 'nearly ideal' claim rests on a comparison to Fig. 7, an idealized collision table whose derivation is never given. That is a real gap, and it matters because the test protocols for the learned controllers (constant/random/reactive traffic) may not be the same as whatever cooperative assumptions generated Fig. 7. The stress-test note is on target here, though I'd phrase it as 'unverified' rather than 'appears to assume' — the paper simply doesn't say.

What the paper does well: it carefully lays out state variables, reward structure, and training details for both the three-vehicle and full-scene settings. The full-scene model that picks the closest vehicles to avoid exponential state growth is a reasonable approach and is described clearly. The reported collision rates for gap sizes >= 15m are low, and the paper honestly notes when performance degrades (e.g., Section 5.1 where the NN settles to extreme acceleration limits, which the authors explicitly say is not ideal performance).

The soft spots, in order of severity. First, Fig. 7. Without knowing the dynamics, number of vehicles, or action model behind the 'best possible' table, the comparison is unfalsifiable in its current form. The authors need to derive the table or at least specify the exact optimization problem. Second, the best network is selected post hoc from training snapshots based on test performance (Section 4.4). That is selection on the test set; without a held-out split or multiple seeds, the collision percentages are optimistically biased. Third, there are no error bars or test counts, so we cannot judge whether a 0% cell means 0/10 or 0/10,000. Fourth, no code or data is released, which makes it hard to check the ideal table or reproduce the results.

None of these are fatal to the approach. The paper is coherent and the authors are transparent about what they did. But the 'nearly ideal' claim is not yet substantiated with the rigor expected for safety-related autonomous driving research. If the authors can pin down the ideal baseline and add statistical care, this would be a solid contribution.

Who this is for: researchers working on multi-agent RL for highway merging, and reviewers of similar work. It deserves a serious referee — I'd send it out, but with the expectation that the authors will need to supply the missing baseline derivation and more careful evaluation before acceptance.

Recommendation: engage with it, but treat the headline claim as conditional until Fig. 7 is documented and the model selection is handled properly.

Referee Report

4 major / 4 minor

Summary. The paper presents a systematic simulation study of multi-agent deep reinforcement learning (DDPG) for longitudinal control during autonomous highway ramp merging. It extends the authors' previous two-vehicle work to a three-vehicle scene (one merging vehicle, two traffic vehicles) and a full scene (two merging vehicles, two or more traffic vehicles). The central claim, stated in the abstract and Section 1, is that the learned controllers are 'nearly ideal when measured against idealized optimal controllers,' supported mainly by collision-percentage tables (Figs. 6 and 9) compared to an 'ideal collision table' (Fig. 7). The paper also describes training details, best-controller selection, and observations about training instability.

Significance. If the near-optimality claim were properly established, this would be a valuable systematic study of scaling multi-agent DRL to highway merging with more than two vehicles, with practical implications for autonomous driving verification. The paper's strengths include a clearly described physical scenario, a systematic progression from three-vehicle to full-scene models, and honest reporting of training instability and controller limitations. However, the central quantitative claim currently rests on an unspecified idealized baseline whose protocol does not match the evaluation protocol, and on post hoc model selection without statistical error bars. These issues must be resolved before the claimed near-optimality can be accepted. The paper does not provide machine-checked proofs or parameter-free derivations; its contribution is empirical.

major comments (4)

[Section 4.5, Fig. 7] The near-optimality claim is quantified by comparing collision rates with Fig. 7, the 'ideal collision table of best possible performance,' but the paper never states how this table is computed: no dynamics, no number of vehicles, no traffic-action model, and no derivation are provided. The caption says marked cells are unavoidable 'when the merge and traffic vehicles are acting using the most ideal action selection to avoid a collision,' which describes a cooperative/optimal-control baseline. In contrast, the learned controllers are tested against traffic vehicles following constant-speed, random, or reactive policies (Sections 4.5 and 5.2). If the ideal table assumes a single traffic vehicle that can choose ideal avoidance actions, it is not an achievable upper bound for the stochastic, non-cooperative test distribution. A learned controller could therefore appear 'nearly ideal' simply because the baseline assumes away the very behaviors used in evaluation. Because the central claim is quantitative, an unreported benchmark makes it unfalsifiable; the authors must either derive Fig. 7 from specified kinematics and policies, or re-benchmark under a same-protocol lower bound.
[Section 4.4 and Section 5.1] The best network is selected post hoc from training snapshots based on test performance (Section 4.4; the 'best network is at 350K episodes' for three-vehicle and at 4.325M episodes for the full scene in Section 5.1). This selection procedure uses the same test distribution for both selection and final evaluation, which can optimistically bias reported collision rates unless a separate validation set or seed-level replication is used. Additionally, all collision percentages in Figs. 6 and 9 are point estimates without error bars or multiple-seed standard deviations, making it impossible to assess whether the observed differences between conditions are significant. The paper should report confidence intervals or multiple independent training runs.
[Section 5.1 and Fig. 8] The paper explicitly observes that after about 7M episodes the neural network settles to output only extreme acceleration limits (either -5 m/s^2 or 4 m/s^2), and states that 'the only way to achieve ideal performance is to have both acceleration and deceleration actions. Therefore, the settling behavior that occurs is not a learned ideal performance.' This is in direct tension with the abstract's 'nearly ideal' claim: for a substantial portion of training, the learned controller provably cannot match the ideal behavior described by Fig. 7. If the best network at 4.325M episodes avoids this settling, the paper should show its action distribution to support the claim; otherwise the near-optimality conclusion is contradicted by the authors' own observation.
[Section 5.2 and Fig. 9] Figure 9 and the text report full-scene test results only for constant and random traffic policies, not for the reactive policy that was part of the three-vehicle evaluation and part of the training mixture. Since the abstract claims robust performance against varied traffic behaviors, the absence of reactive-policy results for the full scene weakens the robustness claim. The paper should either provide those results or clearly delimit the claim to constant and random policies.

minor comments (4)

[Introduction] In the first sentence, 'A Vs' should be 'AVs' (minor typographical error).
[Section 4.4] The sentence 'When it beings to select both acceleration and deceleration actions' contains a typo: 'beings' should be 'begins.'
[Table 1] The table formatting for the 'Range' column is inconsistent (some entries lack units, e.g., the proximity row shows '{−1, 1} unitless' while other rows have units). Please unify the formatting.
[Section 5.2] The description of the constant policy's behavior at the TIV threshold is slightly ambiguous: 'The speed only decreases when a rear traffic vehicle is below the min TIV threshold of 0.8s' could be read as the vehicle reacting to a rear vehicle, whereas the context suggests the vehicle reacts to a front vehicle. Please clarify.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical DRL evaluation does not reduce by the paper's equations to its ideal baseline or to fitted quantities.

full rationale

The paper's central 'nearly ideal' claim is an empirical comparison between collision frequencies of trained DRL controllers (Figs. 6 and 9) and an idealized collision table (Fig. 7). The ideal table is not constructed from the trained network parameters, reward values, or test outputs; it is an independent, though under-documented, benchmark describing unavoidable collisions under idealized action selection. No equation in the paper makes a tested collision rate equal to a fitted quantity or to that table by construction. The extensive references to the authors' prior work (Schester and Ortiz [30]) carry over the two-vehicle reward function, state parameters, and training protocol; the cited impossibility result is context for why 'nearly ideal' is a meaningful target, and that prior work is a published, externally checkable source rather than an unverified self-citation chain. The main weakness—that Fig. 7's computation is unspecified and may assume cooperative ideal action selection while the test policies are constant, random, or reactive—is a soundness and reproducibility concern about the benchmark, not a circular-derivation concern. Likewise, the paper's own admission that the full-scene NN settles to extreme acceleration limits (Section 5.1) is a performance inconsistency, not evidence that the evaluation reduces to its inputs. Therefore no circular step is exhibited.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a set of hand-chosen reward weights, hyperparameters, traffic settings, and a benchmark table that is not fully specified. These choices shape the learned behavior and the reported collision rates.

free parameters (5)

Reward coefficients = success: 1e3; at-fault collision: -1e5; no-fault collision: -1e6
Hand-chosen to heavily penalize collisions and shape training; no optimization or external justification, and they directly influence the learned policy.
Exploration decay rate = 0.999995
Selected as the best of three tested values (0.9995, 0.99995, 0.999995) based on training reward; cross-validation explicitly deferred to future work (Section 4.3).
Hyperparameters (gamma, learning rate, batch size, memory, hidden width) = gamma=0.9, lr=0.001, batch=32, memory=10K, width=30, layers=2
Set to defaults or prior experience; not systematically tuned, and the paper notes model selection could improve performance (Section 6).
TIV threshold for reactive traffic policy = 0.8 s
Chosen as an aggressive test setting; most drivers prefer gaps >= 0.6 s, so this is an assumption about traffic behavior (Section 4.3).
State range clipping values = gap [-2.5,30] m, speed [-10,10] m/s, max gap 100 m
Hand-set normalization limits in Table 1; clipping can change the effective state representation and affect what the NN learns.

assumptions (4)

domain assumption DDPG with self-play converges to a good policy in this stochastic game.
No convergence guarantee is provided; the paper relies on empirical training curves and selects the best checkpoint.
domain assumption The ideal collision table (Fig. 7) is a valid upper bound for the test conditions.
The computation is not specified; if the ideal actions assume information or cooperation unavailable to the learned controllers, the comparison is unfair.
domain assumption Point-mass longitudinal dynamics with acceleration in [-5,4] m/s^2 adequately captures the merge scenario.
The model ignores lateral control, lane changes, vehicle dimensions beyond point lengths, and sensor noise (Section 4, 6).
domain assumption The reactive traffic policy learned with one-third of the training data represents plausible opponent behavior.
No validation against real human or traffic data is provided (Section 4.3).

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Systematic Study of Multi-Agent Deep Reinforcement Learning for Safe and Robust Autonomous Highway Ramp Entry." pith.science (2026). https://pith.science/paper/G55WXWR6

@misc{pith2026241114593,
  author       = {Pith},
  title        = {Pith review of: A Systematic Study of Multi-Agent Deep Reinforcement Learning for Safe and Robust Autonomous Highway Ramp Entry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G55WXWR6}},
  note         = {Machine review of arXiv:2411.14593}
}

read the original abstract

Vehicles today can drive themselves on highways and driverless robotaxis operate in major cities, with more sophisticated levels of autonomous driving expected to be available and become more common in the future. Yet, technically speaking, so-called "Level 5" (L5) operation, corresponding to full autonomy, has not been achieved. For that to happen, functions such as fully autonomous highway ramp entry must be available, and provide provably safe, and reliably robust behavior to enable full autonomy. We present a systematic study of a highway ramp function that controls the vehicles forward-moving actions to minimize collisions with the stream of highway traffic into which a merging (ego) vehicle enters. We take a game-theoretic multi-agent (MA) approach to this problem and study the use of controllers based on deep reinforcement learning (DRL). The virtual environment of the MA DRL uses self-play with simulated data where merging vehicles safely learn to control longitudinal position during a taper-type merge. The work presented in this paper extends existing work by studying the interaction of more than two vehicles (agents) and does so by systematically expanding the road scene with additional traffic and ego vehicles. While previous work on the two-vehicle setting established that collision-free controllers are theoretically impossible in fully decentralized, non-coordinated environments, we empirically show that controllers learned using our approach are nearly ideal when measured against idealized optimal controllers.

Figures

Figures reproduced from arXiv: 2411.14593 by the authors.

**Figure 2.** Full-Scene Highway Merge: Two vehicles in the [PITH_FULL_IMAGE:figures/full_fig_p001_2.png] view at source ↗

**Figure 3.** Three-vehicle actor-critic NN diagrams. NN diagram. DRL NN training occurs at each step of every episode. There are roughly 50 steps on average per episode. The NN trains to 2.5𝑀 episodes, but the best performer is at 350𝐾, or about 17.5𝑀 training steps. However, the reward assignment for a collision or successful merge occurs only once per episode. The input to the actor-critic DDPG NNs are the vehicle state variab… view at source ↗

Figures from the paper (6 more)

**Figure 5.** Figure 5: Standard test performance of the three-vehicle sce [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 4.** Figure 4: Top: Cumulative reward average of four separate [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 6.** Figure 6: Three-vehicle collision test table. The vertical axis is the ramp length. The horizontal axis shows the starting position [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

**Figure 7.** Figure 7: Ideal collision table of best possible performance [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

**Figure 8.** Figure 8: Training Graph of Moving Mean and Cum. Avg. for Rewards. Near [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

**Figure 9.** Figure 9: Full-Scene Collision Test Table. Constant and random policies standard test table with multiple gap settings of [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 33 canonical work pages

[1]

Fuad Fahruddin Achmad Ali Fikri, Syamsul Arifin

M. Fuad Fahruddin Achmad Ali Fikri, Syamsul Arifin. 2022. The Manual on Uniform Traffic Control Devices (MUTCD) . Technical Report Revision 3. Federal Highway Administration (FHWA). https://mutcd.fhwa.dot.gov

work page 2022
[2]

Keith Bradsher. 2024. China Is Testing More Driverless Cars Than Any Other Country. The New York Times. Retrieved 2024-27-09 from https://www.nytimes. com/2024/06/13/business/china-driverless-cars.html

work page 2024
[3]

Max Chafkin. 2022. Even After $100 Billion, Self-Driving Cars Are Going Nowhere. Bloomberg. Retrieved 2024-23-06 from https://www.bloomberg.com/news/features/2022-10-06/even-after100- billion-self-driving-cars-are-going-nowhere

work page 2022
[4]

Caroline Claus and Craig Boutilier. 1998. The dynamics of reinforcement learning in cooperative multiagent systems. AAAI/IAAI 1998 (1998), 746–752

work page 1998
[5]

Cruise. 2023. One Million Driverless Miles. getcruise.com. Retrieved February 22, 2023 from https://getcruise.com/one-million-driverless-miles/

work page 2023
[6]

Missy Cummings, ME Noorman, and M Verdiccio. 2024. Identifying AI Haz- ards and Responsibility Gaps. In Computer Ethics Across Disciplines: APPLYING Deborah Johnson’s Philosophy To Algorithmic Accountability And Ai . Springer Nature

work page 2024
[7]

Wes Davis. 2024. Waymo’s robotaxi depot is still honking its San Francisco neighbors awake. The Verge. Retrieved 2024-27-09 from https://www.theverge.com/2024/8/18/24223160/waymo-honking-san- francisco-parking-lot-depot-fix-not-working

work page 2024
[8]

Chiyu Dong, John M Dolan, and Bakhtiar Litkouhi. 2017. Interactive ramp merging planning in autonomous driving: Multi-Merging leading PGM (MML- PGM). In 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 1–6

work page 2017

Show all 42 references

[9]

Dolan, and Bakhtiar Litkouhi

Chiyu Dong, John M. Dolan, and Bakhtiar Litkouhi. 2018. Smooth behavioral estimation for ramp merging control in autonomous driving. In IEEE Intelligent Vehicles Symposium, Proceedings, Vol. 2018-June. 1692–1697. https://doi.org/10. 1109/IVS.2018.8500576

2018
[10]

Shuo Feng, Haowei Sun, Xintao Yan, Haojie Zhu, Zhengxia Zou, Shengyin Shen, and Henry X. Liu. 2023. Dense reinforcement learning for safety validation of autonomous vehicles. Nature 615, 7953 (2023), 620–627. https://doi.org/10.1038/ s41586-023-05732-2

2023
[11]

Ford. 2024. Ford BlueCruise. Retrieved 2024-10-13 from https://www.ford.com/ technology/bluecruise

2024
[12]

General Motors. 2024. Super Cruise Driver Assistance - Hands-Free Driving. Re- trieved 2024-10-13 from https://www.gmc.com/connectivity-technology/super- cruise

2024
[13]

Alyssa Goard. 2023. Cruise cars cause traffic jam in San Francisco’s North Beach neighborhood. NBC Bay Area. Retrieved 2024-27-09 from https://www.nbcbayarea.com/news/local/cruise-cars-standstill-traffic-san- francisco-north-beach/3294264/

2023
[14]

Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-Term Memory. Neural Computation 9, 8 (11 1997), 1735–1780. https: //doi.org/10.1162/neco.1997.9.8.1735 arXiv:https://direct.mit.edu/neco/article- pdf/9/8/1735/813796/neco.1997.9.8.1735.pdf

1997 doi
[15]

Danial Kamran, Yu Ren, and Martin Lauer. 2021. High-level Decisions from a Safe Maneuver Catalog with Reinforcement Learning for Safe and Cooperative Automated Merging. IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC 2021-Septe (2021), 804–811. https:/...

2021
[16]

Lora Kolodny. 2023. Cruise confirms robotaxis rely on human as- sistance every four to five miles. CNBC. Retrieved 2024-23- 06 from https://www.cnbc.com/2023/11/06/cruise-confirms-robotaxis-rely-on- human-assistance-every-4-to-5-miles.html

2023
[17]

Philip Koopman and William Widen. 2024. Redefining Safety for Autonomous Vehicles. In International Conference on Computer Safety, Reliability, and Security . Springer, 300–314

2024
[18]

Philip J. Koopman. 2022. How Safe Is Safe Enough?: Measuring and Predicting Autonomous Vehicle Safety. Independently published. 368 pages

2022
[19]

Kirsten Korosec. 2023. Ford, VW-backed Argo Ai is shutting down. TechCrunch. https://techcrunch.com/2022/10/26/ford-vw-backed-argo-ai-is- shutting-down/

2023
[20]

Lillicrap, Jonathan J

Timothy P. Lillicrap, Jonathan J. Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, and Daan Wierstra. 2015. Continuous control with deep reinforcement learning. In ICLR 2016. arXiv:1509.02971 http://arxiv. org/abs/1509.02971

2015 arXiv
[21]

Boudette

Cade Metz and Neal E. Boudette. 2021. Inside Tesla as Elon Musk Pushed an Unflinching Vision for Self-Driving Cars. NY Times. Retrieved 2021-12- 13 from https://www.nytimes.com/2021/12/06/technology/tesla-autopilot-elon- musk.html

2021
[22]

Cade Metz, Jason Henry, Ben Laffin, Rebecca Lieberman, and Yiwen Lu. 2024. How Self-Driving Cars Get Help From Humans Hundreds of Miles Away. The New York Times. Retrieved 2024-27-09 from https://www.nytimes.com/interactive/2024/09/03/technology/zoox-self- driving-cars-remote-...

2024
[23]

Volodymyr Mnih, Adria Puigdomenech Badia, Mehdi Mirza, Alex Graves, Tim- othy Lillicrap, Tim Harley, David Silver, and Koray Kavukcuoglu. 2016. Asyn- chronous Methods for Deep Reinforcement Learning. In Proceedings of The 33rd International Conference on Machine Learning (Proc...

2016
[24]

Christopher Nowakowski, Steven E Shladover, Delphine Cody, Fanping Bu, Jes- sica O’Connell, John Spring, Susan Dickey, and David Nelson. 2011. Cooperative adaptive cruise control: Testing drivers’ choices of following distances . Technical Report California PATH Research Repor...

2011 doi
[25]

Kate O’Keeffe and Heather Somerville. 2022. Self-Driving Truck Accident Draws Attention to Safety at TuSimple. The Wall Street Journal. Retrieved 2024- 23-06 from https://www.wsj.com/articles/self-driving-truck-accident-draws- attention-to-safety-at-tusimple-11659346202

2022
[26]

Praveena Penmetsa, Emmanuel Kofi Adanu, Dustin Wood, Teng Wang, and Steven L. Jones. 2019. Perceptions and expectations of autonomous vehicles – A snapshot of vulnerable road user opinion. Technological Forecasting and Social Change 143, March (2019), 9–13. https://doi.org/10....

2019 doi
[27]

Abhirup Roy. 2024. How GM’s Cruise robotaxi tech failures led it to drag pedestrian 20 feet. Reuters. Retrieved 2024-23-06 from https://www.reuters.com/business/autos-transportation/how-gms-cruise- robotaxi-tech-failures-led-it-drag-pedestrian-20-feet-2024-01-26/

2024
[28]

SAE International. 2020. Safety-Relevant Guidance for On-Road Testing of Pro- totype Automated Driving System (ADS)-Operated Vehicles . Technical Report. On-Road Automated Driving (ORAD) Committee. https://doi.org/10.4271/J3018_ 202012

2020 doi
[29]

Larry Schester. 2023. Multi-Agent Reinforcement Learning Autonomous Driving Highway On-Ramp Merge. Ph.D. Thesis. University of Michigan-Dearborn, De- partment of Electrical and Computer Engineering, College of Engineering and Computer Science

2023
[30]

Larry Schester and Luis E. Ortiz. 2021. Automated driving highway traffic merging using deep multi-agent reinforcement learning in continuous state- action spaces. In IEEE Intelligent Vehicles Symposium, Proceedings . Institute of Electrical and Electronics Engineers Inc., 280...

2021
[31]

Shai Shalev-Shwartz, Shaked Shammah, and Amnon Shashua. 2016. Safe, multi-agent, reinforcement learning for autonomous driving. arXiv preprint arXiv:1610.03295 (2016). https://doi.org/10.5626/ktcp.2018.24.12.670 arXiv:arXiv:1610.03295v1

2016 arXiv
[32]

Yichuan Charlie Tang. 2019. Towards Learning Multi-Agent Negotiations via Self-Play. In 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW). 2427–2435. https://doi.org/10.1109/ICCVW.2019.00297

2019
[33]

Brad Templeton. 2023. California DMV Shuts Down Cruise Robotaxis Due To Dragging Of Pedestrian. Forbes. Retrieved 2024-23-06 from https://www.forbes.com/sites/bradtempleton/2023/10/24/california-dmv- shuts-down-cruise-robotaxis-due-to-dragging-of-pedestrian/

2023
[34]

Tesla. 2024. Autopilot and Full Self-Driving (Supervised). Retrieved 2024-10-13 from https://www.tesla.com/support/autopilot

2024
[35]

Trisha Thadani. 2024. Multiple Tesla lawsuits test claim drivers are responsible in Autopilot crashes. Washington Post. Retrieved 2024-23-06 from https://www. washingtonpost.com/technology/2024/04/28/tesla-trial-autopilot-lawsuit/

2024
[36]

Shivansh Tiwary. 2024. Aptiv to cut stake in robotaxi tech firm Motional, stop funding. Reuters. https://www.reuters.com/business/autos-transportation/aptiv- beats-quarterly-profit-estimates-strong-demand-auto-parts-2024-01-31/

2024
[37]

Samuel Triest, Adam Villaflor, and John M. Dolan. 2020. Learning Highway Ramp Merging Via Reinforcement Learning with Temporally-Extended Actions. InIEEE Intelligent Vehicles Symposium, Proceedings. Institute of Electrical and Electronics Engineers Inc., 1595–1600. https://doi...

2020
[38]

Trent Victor, Kristofer Kusano, Tilia Gode, Ruoshu Chen, and Matthew Schwall
[39]

Junqing Wei, John M Dolan, and Bakhtiar Litkouhi. 2013. Autonomous vehicle social behavior for highway entrance ramp management. In 2013 IEEE Intelligent Vehicles Symposium (IV). 201–207

2013
[40]

Zeyi Yang. 2024. How China is regulating robotaxis. MIT Technology Re- view. Retrieved 2024-27-09 from https://www.technologyreview.com/2024/01/ 24/1086989/china-regulation-robotaxi-autonomous-driving/

2024
[41]

Yong Yu, Xiaosheng Si, Changhua Hu, and Jianxun Zhang. 2019. A Review of Recurrent Neural Networks: LSTM Cells and Net- work Architectures. Neural Computation 31, 7 (07 2019), 1235–1270. https://doi.org/10.1162/neco_a_01199 arXiv:https://direct.mit.edu/neco/article- pdf/31/7/1...

2019 doi
[2017]

(2017), 1–30

Safety Performance of the Waymo Rider-Only Automated Driving System at One Million Miles. (2017), 1–30

2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Fuad Fahruddin Achmad Ali Fikri, Syamsul Arifin

M. Fuad Fahruddin Achmad Ali Fikri, Syamsul Arifin. 2022. The Manual on Uniform Traffic Control Devices (MUTCD) . Technical Report Revision 3. Federal Highway Administration (FHWA). https://mutcd.fhwa.dot.gov

work page 2022

[2] [2]

Keith Bradsher. 2024. China Is Testing More Driverless Cars Than Any Other Country. The New York Times. Retrieved 2024-27-09 from https://www.nytimes. com/2024/06/13/business/china-driverless-cars.html

work page 2024

[3] [3]

Max Chafkin. 2022. Even After $100 Billion, Self-Driving Cars Are Going Nowhere. Bloomberg. Retrieved 2024-23-06 from https://www.bloomberg.com/news/features/2022-10-06/even-after100- billion-self-driving-cars-are-going-nowhere

work page 2022

[4] [4]

Caroline Claus and Craig Boutilier. 1998. The dynamics of reinforcement learning in cooperative multiagent systems. AAAI/IAAI 1998 (1998), 746–752

work page 1998

[5] [5]

Cruise. 2023. One Million Driverless Miles. getcruise.com. Retrieved February 22, 2023 from https://getcruise.com/one-million-driverless-miles/

work page 2023

[6] [6]

Missy Cummings, ME Noorman, and M Verdiccio. 2024. Identifying AI Haz- ards and Responsibility Gaps. In Computer Ethics Across Disciplines: APPLYING Deborah Johnson’s Philosophy To Algorithmic Accountability And Ai . Springer Nature

work page 2024

[7] [7]

Wes Davis. 2024. Waymo’s robotaxi depot is still honking its San Francisco neighbors awake. The Verge. Retrieved 2024-27-09 from https://www.theverge.com/2024/8/18/24223160/waymo-honking-san- francisco-parking-lot-depot-fix-not-working

work page 2024

[8] [8]

Chiyu Dong, John M Dolan, and Bakhtiar Litkouhi. 2017. Interactive ramp merging planning in autonomous driving: Multi-Merging leading PGM (MML- PGM). In 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 1–6

work page 2017