mlsanityData to 14 Oct 2026

Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning

arXiv 2609.30258 ↗First posted 24 Sep 2026cs.LGcited by 0

On the map: Deep reinforcement learning methods → RL for energy, networks and operations

Abstract (arXiv)

Distributed learning in embodied reinforcement-learning agents offers a degree of privacy by retaining raw sensor data on-device and transmitting only policy gradients to the server. Yet temporal structure can amplify this leakage beyond single-frame attacks. We introduce Temporal Reconstruction Attack on Consecutive Encodings (TRACE), an amortized temporal gradient-inversion attack that autoregressively reconstructs the sequence of private observation-action trajectories from per-step policy-learning gradients. The attack exploits two structural signals ignored by prior single-frame methods: (i) cross-time correlation between successive embodied gradients, which we formalize via a conditional mutual-information bound, and (ii) closed-form action recovery from policy-head gradient structure, which we prove exact when standard entropy regularization is sufficiently small. On held-out embodied scenes, TRACE reaches 18.8 dB PSNR with near-perfect action recovery at 3-4.5 ms per reconstructed frame, dominating the learning-based baseline across all reconstruction metrics and exceeding optimization attacks while running orders of magnitude faster. Further evaluation demonstrates TRACE's broader applicability across recurrent, residual, and compact transformer victim architectures, multi-modal inputs, and larger discrete action spaces. Defense experiments suggest that protecting temporal gradient streams may require sequence-aware privacy mechanisms.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in RL for energy, networks and operations, one of 15 directions in Deep reinforcement learning methods. Builds on 38 of its 46 references that are in the map. First posted on arXiv 24 Sep 2026.

Builds on

38 of 46 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025Gradient Inversion in Federated Reinforcement Learning29 Nov

Other references · 37

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  3. 2017Proximal Policy Optimization Algorithmscited by 15,215
  4. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  5. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,703
  6. 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
  7. 2016Deep Learning with Differential Privacycited by 2,728
  8. 2016Asynchronous Methods for Deep Reinforcement Learningcited by 2,258
  9. 2015High-Dimensional Continuous Control Using Generalized Advantage Estimationcited by 2,026
  10. 2019Deep Leakage from Gradientscited by 1,021
  11. 2018IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architecturescited by 883
  12. 2017AI2-THOR: An Interactive 3D Environment for Visual AIcited by 869
25 more
  1. 2015Scheduled Sampling for Sequence Prediction with Recurrent Neural Networkscited by 803
  2. 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 712
  3. 2015Sequence Level Training with Recurrent Neural Networkscited by 576
  4. 2017Differentially Private Federated Learning: A Client Level Perspectivecited by 459
  5. 2020iDLG: Improved Deep Leakage from Gradientscited by 382
  6. 2021See through Gradients: Image Batch Recovery via GradInversioncited by 326
  7. 2017Sparse Communication for Distributed Gradient Descentcited by 249
  8. 2021Evaluating Gradient Inversion Attacks and Defenses in Federated Learningcited by 196
  9. 2021Robbing the Fed: Directly Obtaining Private Data in Federated Learning with Modified Modelscited by 108
  10. 2021Gradient Inversion with Generative Image Priorcited by 101
  11. 2020Provable Defense against Privacy Leakage in Federated Learning from Representation Perspectivecited by 94
  12. 2022Fishing for User Data in Large-Batch Federated Learning via Gradient Magnificationcited by 73
  13. 2022Auditing Privacy Defenses in Federated Learning via Generative Gradient Leakagecited by 72
  14. 2021Eluding Secure Aggregation in Federated Learning via Model Inconsistencycited by 67
  15. 2022GradViT: Gradient Inversion of Vision Transformerscited by 53
  16. 2021Gradient-Leakage Resilient Federated Learningcited by 40
  17. 2023LOKI: Large-scale Data Reconstruction Attack against Federated Learning through Model Manipulationcited by 36
  18. 2022Learning to Invert: Simple Adaptive Attacks for Gradient Inversion in Federated Learningcited by 30
  19. 2025Taming Teacher Forcing for Masked Autoregressive Video Generationcited by 30
  20. 2021APRIL: Finding the Achilles' Heel on Privacy for Vision Transformerscited by 24
  21. 2023Scale-MIA: A Scalable Model Inversion Attack against Secure Federated Learning via Latent Space Reconstructioncited by 21
  22. 2023ORL-AUDITOR: Dataset Auditing in Offline Deep Reinforcement Learningcited by 11
  23. 2025CENSOR: Defense Against Gradient Inversion via Orthogonal Subspace Bayesian Samplingcited by 5
  24. 2023Privacy Risks in Reinforcement Learning for Household Robotscited by 4
  25. 2025SVDefense: Effective Defense against Gradient Inversion Attacks via Singular Value Decompositioncited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Privileged observations enable rapid and reliable policy discovery directly in the physical world
  2. 2026Security and Privacy in Large-Model-Driven Embodied Agents: Attacks, Defenses, and Future Directions
  3. 2026Test-time Adversarial Takeover: A Real-time Hijacking Interface against Robotic Diffusion Policies
  4. 2024Streaming Diffusion Policy: Fast Policy Synthesis with Variable Noise Diffusion Models
  5. 2026TOUR: A Trajectory-Level Unlearning Benchmark for Offline Reinforcement Learning
  6. 2026Security of Foundation-Model-Powered Embodied Agents: Attack Surfaces, Attacks, Defenses, and Evaluation
  7. 2025TS-DP: Reinforcement Speculative Decoding For Temporal Adaptive Diffusion Policy Acceleration
  8. 2021XIRL: Cross-embodiment Inverse Reinforcement Learning