Robust Deep Reinforcement Learning in Robotics via Adaptive Gradient-Masked Adversarial Attacks

arXiv 2503.20844 ↗First posted 26 Mar 2025cs.LG cs.AI cs.NIcited by 7

Not on the map. Why?

Abstract (arXiv)

Deep reinforcement learning (DRL) has emerged as a promising approach for robotic control, but its realworld deployment remains challenging due to its vulnerability to environmental perturbations. Existing white-box adversarial attack methods, adapted from supervised learning, fail to effectively target DRL agents as they overlook temporal dynamics and indiscriminately perturb all state dimensions, limiting their impact on long-term rewards. To address these challenges, we propose the Adaptive Gradient-Masked Reinforcement (AGMR) Attack, a white-box attack method that combines DRL with a gradient-based soft masking mechanism to dynamically identify critical state dimensions and optimize adversarial policies. AGMR selectively allocates perturbations to the most impactful state features and incorporates a dynamic adjustment mechanism to balance exploration and exploitation during training. Extensive experiments demonstrate that AGMR outperforms state-of-the-art adversarial attack methods in degrading the performance of the victim agent and enhances the victim agent's robustness through adversarial defense mechanisms.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Builds on 51 of its 77 references that are in the map. First posted on arXiv 26 Mar 2025.

Builds on

51 of 77 references.

References · 51

2017Proximal Policy Optimization Algorithmscited by 14,441
2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
2014Explaining and Harnessing Adversarial Examplescited by 7,287
2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
2013Playing Atari with Deep Reinforcement Learningcited by 3,211
2015High-Dimensional Continuous Control Using Generalized Advantage Estimationcited by 1,916
2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
2017Constrained Policy Optimizationcited by 928
2017Ensemble Adversarial Training: Attacks and Defensescited by 833
2020Fast is better than free: Revisiting adversarial trainingcited by 723
2018Improving Transferability of Adversarial Examples with Input Diversitycited by 591
2019Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attackscited by 440

38 more

2019Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attackscited by 352
2017Adversarial Attacks on Neural Network Policiescited by 314
2018Adversarial Attacks and Defences: A Surveycited by 259
2020Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observationscited by 235
2017Tactics of Adversarial Attack on Deep Reinforcement Learning Agentscited by 163
2021Admix: Enhancing the Transferability of Adversarial Attackscited by 155
2017Robust Deep Reinforcement Learning with Adversarial Attackscited by 152
2023Safety-Gymnasium: A Unified Safe Reinforcement Learning Benchmarkcited by 135
2021Advances in adversarial attacks and defenses in computer vision: A surveycited by 127
2020SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulationcited by 118
2023Efficient Parallel Split Learning over Resource-constrained Wireless Edge Networkscited by 95
2023Diffusion Models for Imperceptible and Transferable Adversarial Attackcited by 93
2022Distributionally Robust Model-Based Offline Reinforcement Learning with Near-Optimal Sample Complexitycited by 85
2020Robust Deep Reinforcement Learning through Adversarial Losscited by 73
2023Content-based Unrestricted Adversarial Attackcited by 73
2018Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Networkcited by 68
2024AdaptSFL: Adaptive Split Federated Learning in Resource-constrained Edge Networkscited by 64
2020Challenges and Countermeasures for Adversarial Attacks on Deep Reinforcement Learningcited by 61
2024SplitLoRA: A Split Parameter-Efficient Fine-Tuning Framework for Large Language Modelscited by 60
2021AdvDrop: Adversarial Attack to DNNs by Dropping Informationcited by 55
2024Automated Federated Pipeline for Parameter-Efficient Fine-Tuning of Large Language Modelscited by 49
2024AgentsCoDriver: Large Language Model Empowered Collaborative Driving with Lifelong Learningcited by 40
2023FedSN: A Federated Learning Framework over Heterogeneous LEO Satellite Networkscited by 38
2024Hierarchical Split Federated Learning: Convergence Analysis and System Optimizationcited by 38
2021Exploring Misclassifications of Robust Neural Networks to Enhance Adversarial Attackscited by 33
2023Optimal Resource Allocation for U-Shaped Parallel Split Learningcited by 33
2024FedAC: An Adaptive Clustered Federated Learning Framework for Heterogeneous Datacited by 25
2025A Survey on Backdoor Threats in Large Language Models (LLMs): Attacks, Defenses, and Evaluationscited by 25
2024AgentsCoMerge: Large Language Model Empowered Collaborative Decision Making for Ramp Mergingcited by 22
2025LEO-Split: A Semi-Supervised Split Learning Framework over LEO Satellite Networkscited by 22
2024SatFed: A Resource-Efficient LEO Satellite-Assisted Heterogeneous Federated Learning Frameworkcited by 21
2024IC3M: In-Car Multimodal Multi-object Monitoring for Abnormal Status of Both Driver and Passengerscited by 20
2024MERIT: Multimodal Wearable Vital Sign Waveform Monitoringcited by 16
2022Robust Adversarial Attacks Detection based on Explainable Deep Reinforcement Learning For UAV Guidance and Planningcited by 13
2024Robust Deep Reinforcement Learning Through Adversarial Attacks and Training : A Surveycited by 13
2023Towards Building More Robust Models with Frequency Biascited by 12
2025Rethinking Adversarial Attacks in Reinforcement Learning from Policy Distribution Perspectivecited by 10
2025A Contemporary Survey of Large Language Model Assisted Program Analysiscited by 10

50 most cited of 51 All 51 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 5; 2026: 2.

7 papers.