UNDREAM: Bridging Differentiable Rendering and Photorealistic Simulation for End-to-end Adversarial Attacks
arXiv:2510.16923v3 Announce Type: replace-cross Abstract: Deep learning models deployed in safety critical applications like autonomous driving use simulations to test their robustness against adversarial attacks in realistic conditions. However, these simulations are non-differentiable, forcing researchers to create attacks that do not integrate simulation environmental factors, reducing attack success. To address this limitation, we introduce UNDREAM, the first software framework that bridges the gap between photorealistic simulators and differentiable renderers to enable end-to-end optimiza
The increasing deployment of deep learning models in safety-critical autonomous systems necessitates more robust adversarial attack testing, which current non-differentiable simulations do not adequately support.
This development enables more effective and realistic adversarial attacks on AI systems in critical applications, directly impacting their security and reliability in real-world environments.
The ability to run end-to-end optimizations for adversarial attacks by bridging photorealistic simulators with differentiable renderers will significantly improve the efficacy of sophisticated attacks.
- · Adversarial AI researchers
- · Cybersecurity firms specializing in AI red-teaming
- · Developers of robust AI defense mechanisms
- · Developers of vulnerable AI systems
- · Sectors reliant on unhardened autonomous AI
- · Systems lacking integrated adversarial robustness testing
Adversarial attacks on autonomous driving and other safety-critical AI systems become significantly more sophisticated and successful.
This drives a substantial increase in demand for advanced AI security and robustness solutions, leading to new R&D and product cycles.
Public trust in AI-driven autonomous systems may fluctuate as the cat-and-mouse game between attackers and defenders intensifies, potentially impacting adoption rates.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI