Temporal gradient attack reconstructs private trajectories from embodied RL data
Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning
Machine Learning
Summary
Training AI agents that act in the world without sharing their exact sensory experiences helps protect privacy. The authors show it is possible to reverse-engineer these private experiences by analyzing patterns in how the agent’s learning updates change over time. They developed an attack called TRACE that uses the timing and structure of gradients to reconstruct sequences of what the agent saw and did, more accurately and quickly than earlier methods. This raises concerns about privacy in systems that share learning updates instead of raw data.
What this means in practice
- •For ai security teams: Detect and mitigate risks from reconstructing private agent trajectories from shared policy gradients in distributed learning.
- •For robotics engineers: Evaluate privacy vulnerabilities in embodied AI systems that share gradients to improve secure robot deployment.
Authors
Sudip Bhujel, Shanghao Shi, Ruiquan Huang, Ning Zhang, Yang Xiao
Abstract
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.