Haojie Xin

Haojie Xin, Xiaodong Zhang, Renzhi Tang et al.

Simulation is pivotal in evaluating the performance of autonomous driving systems due to the advantages of high efficiency and low cost compared to on-road testing. Bridging the gap between simulation and the real world requires realistic agent behaviors. However, the existing works have the following shortcomings in achieving this goal: (1) log replay offers realistic scenarios but often leads to collisions due to the absence of dynamic interactions, and (2) both heuristic-based and data-based solutions, which are parameterized and trained on real-world datasets, encourage interactions but often deviate from real-world data over long horizons. In this work, we propose LitSim, a long-term interactive simulation approach that maximizes realism by minimizing the interventions in the log. Specifically, our approach primarily uses log replay to ensure realism and intervenes only when necessary to prevent potential conflicts. We then encourage interactions among the agents and resolve the conflicts, thereby reducing the risk of unrealistic behaviors. We train and validate our model on the real-world dataset NGSIM, and the experimental results demonstrate that LitSim outperforms the currently popular approaches in terms of realism and reactivity.

Peiyuan Tang, Haojie Xin, Xiaodong Zhang et al.

As Vision-Language Models (VLMs) demonstrate increasing capabilities across real-world applications such as code generation and chatbot assistance, ensuring their safety has become paramount. Unlike traditional Large Language Models (LLMs), VLMs face unique vulnerabilities due to their multimodal nature, allowing adversaries to modify visual or textual inputs to bypass safety guardrails and trigger the generation of harmful content. Through systematic analysis of VLM behavior under attack, we identify a novel phenomenon termed ``delayed safety awareness''. Specifically, we observe that safety-aligned VLMs may initially be compromised to produce harmful content, but eventually recognize the associated risks and attempt to self-correct. This pattern suggests that VLMs retain their underlying safety awareness but experience a temporal delay in their activation. Building on this insight, we hypothesize that VLMs' safety awareness can be proactively reactivated through carefully designed prompts. To this end, we introduce ``The Safety Reminder'', a soft prompt tuning approach that optimizes learnable prompt tokens, which are periodically injected during the text generation process to enhance safety awareness, effectively preventing harmful content generation. Additionally, our safety reminder only activates when harmful content is detected, leaving normal conversations unaffected and preserving the model's performance on benign tasks. Through comprehensive evaluation across three established safety benchmarks and one adversarial attacks, we demonstrate that our approach significantly reduces attack success rates while maintaining model utility, offering a practical solution for deploying safer VLMs in real-world applications.