Quantum Artificial Intelligence for Secure Autonomous Vehicle Navigation: An Architectural Proposal

Hemanth Kannamarlapudi; Sowmya Chintalapudi

REVIEW 4 major objections 7 minor 17 references

Quantum Artificial Intelligence for Secure Autonomous Vehicle Navigation: An Architectural Proposal

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper proposes a complete quantum AI navigation stack for self-driving cars, in which quantum neural networks fuse multimodal sensors, Nav-Q quantum reinforcement learning selects maneuvers, and post-quantum cryptography…

desk verdict A clearly labeled architectural proposal that integrates known quantum components, but the performance claims rest on unvalidated hardware assumptions and a representational-capacity argument that doesn't deliver what the abstract promises. read the letter →

arxiv 2506.16000 v1 pith:GKGGSMFC submitted 2025-06-19 cs.ET cs.AIcs.ROquant-ph

Hemanth Kannamarlapudi , Sowmya Chintalapudi This is my paper · ORCID

classification cs.ETcs.AIcs.ROquant-ph

keywords quantumartificialintelligenceautonomousvehiclesneuralnetworkssensorfusionreinforcementlearningpost-quantumcryptographyNav-Qadversarialrobustness

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 proposes a full-stack quantum AI architecture for autonomous vehicle navigation. It claims that quantum neural networks can fuse heterogeneous sensors into a single quantum state, that the Nav-Q quantum reinforcement learning module can convert that state into driving decisions, and that post-quantum cryptography can keep the entire loop secure against classical and quantum attacks. A sympathetic reader should read it as a design proposal that promises quantum-level performance and security in one coherent pipeline, rather than as a demonstrated system. The authors themselves list the missing pieces: adequate 50 to 100 qubit hardware, sub-50 ms coherence and latency, barren-plateau-free training, and empirical validation.

What carries the argument

The load-bearing object is the amplitude-encoded quantum state. In the QNN module, each sensor vector $s_i \in \mathbb{R}^{d_i}$ is mapped into a normalized superposition $|\psi\rangle = \frac{1}{\sqrt{N}} \sum_{i=1}^{n} \sum_{j=1}^{d_i} \alpha_{i,j} s_{i,j} |i,j\rangle$, where $\alpha_{i,j}$ are learned attention weights. The trainable variational circuit alternates parameterized rotations and entangling operations, $U(\theta) = \prod_{\ell=1}^{L} [\prod_{q=1}^{Q} R_Y(\theta_{\ell,q})][\prod_{q=1}^{Q-1} \mathrm{CNOT}_{q,q+1}]$, optimized by quantum policy gradient methods. Nav-Q is the named quantum reinforcement learning framework that consumes this fused state and outputs steering, acceleration, and braking decisions; CRYSTALS-Kyber and CRYSTALS-Dilithium provide the post-quantum authenticated channel that protects the loop. This chaining of fusion, policy, and security in one quantum state is what carries the paper's central claim.

What would settle it

A single decisive test would be to train the proposed quantum fusion and Nav-Q policy modules with a public autonomous-driving dataset and compare navigation reward and end-to-end latency against a classical deep-reinforcement-learning baseline using the same data. If the classical baseline matches or beats the quantum pipeline, the central claim of quantum-enhanced navigation is refuted; if current 50 to 100 qubit hardware cannot sustain coherence through one fusion-and-policy cycle within 50 ms, the real-time premise fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that an autonomous vehicle can be navigated through a single integrated quantum AI pipeline rather than separate classical perception, planning, and security modules. The QNN maps LiDAR, radar, camera, GPS, and weather inputs into one quantum state; Nav-Q turns that state into control commands; and post-quantum cryptography authenticates every sensor stream and actuation message. The authors assert that this preserves quantum correlations end to end, yielding better navigation policies under dynamic conditions and security against both classical and future quantum attacks. They also state plainly that this is an architectural and theoretical proposal: current quantum hardware cannot yet meet the real-time requirements, and training and validation remain future work.

Load-bearing premise

The architecture's claimed quantum performance rests on a 50 to 100 qubit processor running the fusion and policy circuits in under 50 milliseconds, with training gradients intact and coherence preserved, at automotive scale; the authors explicitly say current hardware is not there yet and no experiment tests it.

Editorial extensions

If this is right

If the architecture is correct, perception, planning, and security stop being separate modules: driving actions would be derived from one quantum-fused state and authenticated through one control loop.
A 50 to 100 qubit fusion processor would, under the paper's exponential state-space argument, encode cross-sensor combinations that polynomial classical fusion cannot represent.
Extending post-quantum authentication to sensor-to-processor links would close an attack surface left open by V2V-only cryptographic protections.
The adversarial training loop would give a path to robustness against sensor attacks such as GPS jamming, LiDAR spoofing, and camera patches, not just image-classification adversarial examples.
Once the hardware assumptions are met, the framework converts several separate quantum computing research threads into a testable end-to-end navigation system.

Reading between the lines

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

Inference: the $O(2^n)$ state-space count is necessary but not sufficient for a quantum advantage; if measurements or classical preprocessing collapse the fused state before policy learning, the pipeline reduces to classical processing of amplitudes. A simulator study varying the number of measurements and comparing policy reward would expose this.
Inference: the likely bottleneck is the quantum-classical interface, not the circuit ansatz; collecting sensor data, running PQC decryption, and converting controls to classical actuation all interrupt coherence. Profiling where coherence is lost would be a concrete test of feasibility.
Inference: the three-layer pattern of fusion, learned policy, and authenticated actuation would transfer to other real-time safety-critical systems such as drone collision avoidance or robotic surgery, so the architecture's value is not limited to autonomous cars.
Inference: the authors' planned comparison against a classical setup is the right falsification design, and it can begin now on quantum simulators using public autonomous-driving data even before real 50 to 100 qubit hardware exists.

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 position paper, not a result. The integrated QNN-Nav-Q-PQC pipeline is genuinely new as a framing, and the authors correctly point out that intra-vehicle sensor-to-processor links have been underserved relative to V2X. The PQC layer is standard but plausible, and the limitations section is unusually honest: they say hardware isn't there, training is future work, and barren plateaus are a risk. That honesty is a real point in their favor.

The soft spot is the loaded language around 'quantum performance' and 'future-proof security.' The theoretical section compares Hilbert space dimension O(2^n) with classical parameters O(poly(n)), but any classical n-bit system also has 2^n states, so that's not a computational speedup. Equation (1) is amplitude encoding with attention weights; Eq. (2) is a generic hardware-efficient ansatz. Nothing shows these circuits solve sensor fusion or navigation better or faster than classical networks. The paper's own limitations and future work defer the experiments that would support the abstract's central claim. The PQC layer is well-motivated, but it doesn't validate the quantum navigation parts.

The reader's strongest claim is accurate; the weakest assumption—that 50–100 qubit processors run the whole stack in under 50 ms without decay or plateaus—is indeed load-bearing and unresolved. The stress-test note is on target, and I agree with the overall reject verdict.

Who is this for? Someone surveying quantum AI architectures in autonomy might skim it as a roadmap, and the intra-vehicle security gap is worth remembering. But there is no mathematical, empirical, or formal result to engage with, and the citations, while relevant, don't substantiate the performance advantage. It doesn't deserve a full peer-review cycle in its current form. I'd send it back with an invitation to resubmit after running even a small simulation or toy implementation that tests the fusion or QRL claims against a classical baseline. That would make the architectural intuition worth referee time.

Referee Report

4 major / 7 minor

Summary. The paper proposes a three-stage quantum AI architecture for autonomous vehicle navigation: QNN-based multimodal sensor fusion via amplitude encoding, Nav-Q quantum reinforcement learning for policy optimization, and post-quantum cryptography (CRYSTALS-Kyber/Dilithium) for intra-vehicle and V2X communication. It claims that the integrated pipeline provides 'quantum performance and future proof security.' Section V presents a theoretical analysis asserting exponential representational capacity, faster convergence, and provable security; Section VI-B concedes that current hardware cannot meet the real-time requirements, and Section VI-C defers training and validation to future work.

Significance. If the architectural claims were validated, this would be a significant step toward quantum-enabled autonomous driving, and the paper usefully identifies the underexplored intra-vehicle communication gap in PQC deployment. The authors are transparent about hardware limitations and the need for future validation. However, no experiments, simulations, or formal proofs support the central advantage claims; the 'theoretical analysis' is mostly assertion. The contribution is a plausible vision statement, not a demonstrated architecture.

major comments (4)

[Sec. V-A, Eq. (5)] The claim that QNNs provide 'exponential representational capacity' because n qubits encode 2^n states (Eq. 5) conflates state-space cardinality with computational advantage: a classical n-bit register also has 2^n possible states, and representational capacity does not imply that the learning algorithm can exploit it efficiently. The paper provides no evidence that the amplitude encoding in Eq. (1) or the hardware-efficient ansatz in Eq. (2) outperforms classical sensor fusion on any navigation task. Since the abstract's 'quantum performance' rests on this point, it needs at least a concrete computational separation argument or an empirical benchmark.
[Sec. V-B] The assertions that Nav-Q provides 'theoretical guarantees for faster convergence' and that quantum interference 'enables escape from suboptimal policy regions' are stated without proof or a specific theorem from reference [4]. Section VI-C explicitly defers training, simulation, and hardware validation to future work, so the quantum advantage in policy learning is currently unsupported. This is load-bearing for the paper's central claim of superior navigation performance.
[Sec. III-A 6 / Sec. VI-B] The sub-50ms latency claim for the integrated QNN-Nav-Q-PQC loop is load-bearing for real-time autonomous driving, yet no analysis or benchmark is provided; Sec. VI-B concedes that current hardware 'isn't quite there yet' for millisecond response times. In addition, the paper does not account for the cryptographic overhead of Dilithium signature generation/verification and Kyber encapsulation on the control loop, which is known to be nontrivial in vehicular settings. Without a latency budget or simulation, the claim that the framework 'maintains sub-50ms latency requirements' is unsupported.
[Sec. V-C] The statement that the integrated pipeline provides 'provable security against quantum adversaries with computational advantages bounded by Security Level ≥ 2^128 operations (NIST Level 3)' is technically inaccurate: NIST security category 3 corresponds to hardness around AES-192, not 2^128 operations (which is closer to Level 1). More importantly, this is a property of the underlying PQC primitives, not of the integrated QNN-Nav-Q-PQC pipeline; no argument addresses security of the measured classical commands, the quantum-classical interface, or side channels. The 'end-to-end quantum integrity' claim is therefore unsupported.

minor comments (7)

[Sec. II-D / References] The citation to Twardokus et al. is given as [7], but the reference list assigns Twardokus et al. to [2]; the list's [7] is Haneche et al. Please fix the cross-reference.
[Sec. II-A] The text cites Zhou et al. for quantum-enhanced sensor fusion, but reference [3] is 'Quantum advantage in learning from experiments'; the cited work may not match the claimed content.
[Sec. IV-A] The role of the learned attention weights α_i,j in Eq. (1) is unclear: are they trained inside the QNN or precomputed classically? If they are classical parameters, the 'quantum fusion' is an amplitude-encoded classical weighted sum; please clarify.
[Sec. IV-B] The notation L_classical(θ, s) in Eq. (3) is confusing, since the loss is computed from quantum measurement outcomes; consider defining this as a hybrid classical-quantum loss.
[Sec. V-C] The phrase 'with computational advantages bounded by' is ill-posed; the security level is a lower bound on attack complexity, not an upper bound on the adversary's computational advantage.
[Sec. III-A / Fig. 1] The pipeline steps in Fig. 1 are notated as 'Step1, Step 2, ...' in the caption; also the figure would benefit from a clear dataflow diagram with latency annotations.
[Abstract] The phrase 'providing quantum performance and future proof security' overstates the evidence presented; this should be qualified as a proposed architecture whose advantages are not yet demonstrated.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the proposed architecture is an assembly of external components; its quantum-performance claims are unsupported but not circular.

full rationale

This is an architectural proposal with no fitted parameters, no trained models, and no empirical predictions. The three pipeline stages (QNN fusion, Nav-Q reinforcement learning, PQC security) are assembled from external references: Nav-Q is explicitly attributed to reference [4] (Sinha et al.), PQC standards are imported from NIST-lattice schemes, and adversarial training follows standard PGD methodology (Eqs. 3-4). There are no self-citations by the present authors, and no load-bearing argument reduces to a prior-work chain by the same authors. The only equation presented as an advantage, Eq. (5), compares the Hilbert-space dimension O(2^n) with a classical parameter count O(poly(n)); this is a definitional statement about qubit state space, not a derived prediction, and it is better characterized as an overstatement than as circularity. The paper's own Limitations (Sec. VI-B) and Future Work (Sec. VI-C) concede that current hardware cannot meet the real-time latency requirement and that training, simulation, and hardware validation are deferred, which confirms that the central 'quantum performance' claim is an unsupported hypothesis rather than a result forced by its own definitions. No circular step can be exhibited, so the appropriate finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper's central claims rest on the feasibility of training quantum circuits for sensor fusion and navigation, and on the transferability of classical adversarial training and PQC security guarantees to an automotive quantum pipeline. These are unproven domain assumptions, not derived results.

free parameters (3)

alpha_i,j (attention weights) = not fitted (proposed trainable)
Learned attention weights for each sensor component in the amplitude encoding of Eq. 1.
theta_l,q (variational parameters) = not fitted (proposed trainable)
Rotation angles for the variational quantum circuit in Eq. 2.
lambda (robustness coefficient) = not fitted
Balances clean and adversarial policy loss in Eq. 4.

assumptions (4)

domain assumption Quantum amplitude encoding can represent multi-modal sensor data in a superposition state with exponential capacity, and this representation is useful for navigation.
Invoked in Sec. IV-A and Sec. V-A; no proof that this theoretical capacity translates to practical AV fusion.
domain assumption Variational quantum circuits with alternating RY rotations and CNOT entangling layers can learn navigation policies via quantum policy gradients.
Assumed in Sec. IV-A and V-B; no convergence or trainability analysis provided.
domain assumption Adversarial training developed for classical networks transfers to quantum circuits for sensor fusion and RL.
Assumed in Sec. IV-B, Eq. 3-4; no experiments support this transfer.
domain assumption CRYSTALS-Kyber and Dilithium provide the stated security levels (>= 2^128 operations) in the embedded automotive setting.
Stated in Sec. V-C; relies on published security claims and does not address automotive-specific side-channels or latency constraints.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantum Artificial Intelligence for Secure Autonomous Vehicle Navigation: An Architectural Proposal." pith.science (2026). https://pith.science/paper/GKGGSMFC

@misc{pith2026250616000,
  author       = {Pith},
  title        = {Pith review of: Quantum Artificial Intelligence for Secure Autonomous Vehicle Navigation: An Architectural Proposal},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GKGGSMFC}},
  note         = {Machine review of arXiv:2506.16000}
}

read the original abstract

Navigation is a very crucial aspect of autonomous vehicle ecosystem which heavily relies on collecting and processing large amounts of data in various states and taking a confident and safe decision to define the next vehicle maneuver. In this paper, we propose a novel architecture based on Quantum Artificial Intelligence by enabling quantum and AI at various levels of navigation decision making and communication process in Autonomous vehicles : Quantum Neural Networks for multimodal sensor fusion, Nav-Q for Quantum reinforcement learning for navigation policy optimization and finally post-quantum cryptographic protocols for secure communication. Quantum neural networks uses quantum amplitude encoding to fuse data from various sensors like LiDAR, radar, camera, GPS and weather etc., This approach gives a unified quantum state representation between heterogeneous sensor modalities. Nav-Q module processes the fused quantum states through variational quantum circuits to learn optimal navigation policies under swift dynamic and complex conditions. Finally, post quantum cryptographic protocols are used to secure communication channels for both within vehicle communication and V2X (Vehicle to Everything) communications and thus secures the autonomous vehicle communication from both classical and quantum security threats. Thus, the proposed framework addresses fundamental challenges in autonomous vehicles navigation by providing quantum performance and future proof security. Index Terms Quantum Computing, Autonomous Vehicles, Sensor Fusion

Figures

Figures reproduced from arXiv: 2506.16000 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 15 canonical work pages

[4]

Nav-Q: Quantum Deep Reinforcement Learning for Collision-Free Navigation of Self-Driving Cars

A. Sinha, A. Kumar, P. Singh, and R. Sharma, “Nav-Q: Quantum Deep Reinforcement Learning for Collision-Free Navigation of Self-Driving Cars,” arXiv preprint arXiv:2311.12875 , 2023

work page Pith review arXiv 2023
[1]

Sensor fusion for predicting vehicles’ path for collision avoidance systems,

A. Polychronopoulos, M. Tsogas, A. J. Amditis, and L. Andreone, “Sensor fusion for predicting vehicles’ path for collision avoidance systems,” IEEE Transactions on Intelligent Transportation Systems, vol. 8, no. 3, pp. 549–562, Sep. 2007

work page 2007
[2]

When Cryp- tography Needs a Hand: Practical Post-Quantum Authentication for V2V Communications,

G. Twardokus, N. Bindel, H. Rahbari, and S. McCarthy, “When Cryp- tography Needs a Hand: Practical Post-Quantum Authentication for V2V Communications,” Network and Distributed System Security Symposium (NDSS), 2024

work page 2024
[3]

Quantum advantage in learning from experiments,

L. Zhou, S. Wang, M. Ezzeddine, et al., “Quantum advantage in learning from experiments,” Nature Communications, vol. 14, article 4876, 2023

work page 2023
[5]

Towards quantum enhanced adversarial robustness in machine learning

M. T. West, S. M. Erfani, and C. Leckie, “Towards quantum en- hanced adversarial robustness in machine learning,” arXiv preprint arXiv:2306.12688, 2023

work page Pith review arXiv 2023
[6]

nuScenes: A multimodal dataset for autonomous driving,

H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuScenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 11,621–11,631

work page 2020
[7]

Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review,

M. Haneche, E. Boudy, P. Charpentier, and R. Bon ´e, “Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review,” Sensors, vol. 20, no. 15, article 4220, 2020

work page 2020
[8]

Multi-sensor fusion and segmentation for autonomous vehicle multi-object tracking using deep Q networks,

K. Rajagopal, A. Kumar, and S. Sharma, “Multi-sensor fusion and segmentation for autonomous vehicle multi-object tracking using deep Q networks,” Scientific Reports, vol. 14, article 26919, 2024

work page 2024

Show all 17 references

[9]

Quantum Deep Re- inforcement Learning for Robot Navigation Tasks,

H. Hohenfeld, J. Schubert, and E. Altman, “Quantum Deep Re- inforcement Learning for Robot Navigation Tasks,” arXiv preprint arXiv:2202.12180, 2022

2022 arXiv
[10]

Robust quantum- inspired reinforcement learning for robot navigation,

D. Dong, C. Chen, B. Qi, I. R. Petersen, and F. Nori, “Robust quantum- inspired reinforcement learning for robot navigation,”IEEE Transactions on Neural Networks and Learning Systems , vol. 23, no. 12, pp. 1884– 1897, 2012

2012
[11]

Quantum-Enhanced Hybrid Reinforcement Learning Framework for Dynamic Path Planning in Autonomous Systems,

S. Tomar, P. Singh, A. Kumar, and R. Sharma, “Quantum-Enhanced Hybrid Reinforcement Learning Framework for Dynamic Path Planning in Autonomous Systems,” arXiv preprint arXiv:2504.20660 , 2024

2024 arXiv
[12]

Post-quantum cryptography for automotive systems,

T. Fritzmann, G. Sigl, and J. Sep ´ulveda, “Post-quantum cryptography for automotive systems,” Microprocessors and Microsystems , vol. 87, article 104359, 2021

2021
[13]

Quantum- Resistant Cryptography for Automotive Cybersecurity: Implementing Post-Quantum Algorithms to Secure Next-Generation Autonomous and Connected Vehicles,

R. Soundarapandiyan, P. Sivathapandi, and A. Selvaraj, “Quantum- Resistant Cryptography for Automotive Cybersecurity: Implementing Post-Quantum Algorithms to Secure Next-Generation Autonomous and Connected Vehicles,”Cybersecurity and Network Defense Research, vol. 3, no. 2, pp...

2023
[14]

Post-Quantum Secure Over- the-Air Update of Automotive Systems,

S. M. Ricci, M. Gerdes, and T. Loruenser, “Post-Quantum Secure Over- the-Air Update of Automotive Systems,” Cryptology ePrint Archive , Report 2023/965, 2023

2023
[15]

Benchmarking adversarially robust quantum machine learning at scale,

C. West, J. Hayes, and M. Dalton, “Benchmarking adversarially robust quantum machine learning at scale,” Physical Review Research , vol. 5, no. 2, article 023186, 2023

2023
[16]

Adversarial Attacks and Defense Technologies on Autonomous Vehicles: A Review,

K. T. Y . Mahima, M. Ayoob, and G. Poravi, “Adversarial Attacks and Defense Technologies on Autonomous Vehicles: A Review,” Applied Computer Systems, vol. 26, no. 2, pp. 96–106, 2021

2021
[17]

Sensor Fusion in Autonomous Vehicle with Traffic Surveillance Camera System: Detection, Localization, and AI Networking,

J. Chen, Y . Liu, and S. Zhang, “Sensor Fusion in Autonomous Vehicle with Traffic Surveillance Camera System: Detection, Localization, and AI Networking,” PMC Article PMC10055109 , 2023

2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [4]

Nav-Q: Quantum Deep Reinforcement Learning for Collision-Free Navigation of Self-Driving Cars

A. Sinha, A. Kumar, P. Singh, and R. Sharma, “Nav-Q: Quantum Deep Reinforcement Learning for Collision-Free Navigation of Self-Driving Cars,” arXiv preprint arXiv:2311.12875 , 2023

work page Pith review arXiv 2023

[2] [1]

Sensor fusion for predicting vehicles’ path for collision avoidance systems,

A. Polychronopoulos, M. Tsogas, A. J. Amditis, and L. Andreone, “Sensor fusion for predicting vehicles’ path for collision avoidance systems,” IEEE Transactions on Intelligent Transportation Systems, vol. 8, no. 3, pp. 549–562, Sep. 2007

work page 2007

[3] [2]

When Cryp- tography Needs a Hand: Practical Post-Quantum Authentication for V2V Communications,

G. Twardokus, N. Bindel, H. Rahbari, and S. McCarthy, “When Cryp- tography Needs a Hand: Practical Post-Quantum Authentication for V2V Communications,” Network and Distributed System Security Symposium (NDSS), 2024

work page 2024

[4] [3]

Quantum advantage in learning from experiments,

L. Zhou, S. Wang, M. Ezzeddine, et al., “Quantum advantage in learning from experiments,” Nature Communications, vol. 14, article 4876, 2023

work page 2023

[5] [5]

Towards quantum enhanced adversarial robustness in machine learning

M. T. West, S. M. Erfani, and C. Leckie, “Towards quantum en- hanced adversarial robustness in machine learning,” arXiv preprint arXiv:2306.12688, 2023

work page Pith review arXiv 2023

[6] [6]

nuScenes: A multimodal dataset for autonomous driving,

H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuScenes: A multimodal dataset for autonomous driving,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 11,621–11,631

work page 2020

[7] [7]

Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review,

M. Haneche, E. Boudy, P. Charpentier, and R. Bon ´e, “Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review,” Sensors, vol. 20, no. 15, article 4220, 2020

work page 2020

[8] [8]

Multi-sensor fusion and segmentation for autonomous vehicle multi-object tracking using deep Q networks,

K. Rajagopal, A. Kumar, and S. Sharma, “Multi-sensor fusion and segmentation for autonomous vehicle multi-object tracking using deep Q networks,” Scientific Reports, vol. 14, article 26919, 2024

work page 2024