Ye Tian

Yuewen Mei, Tong Nie, Jian Sun et al.

Ensuring and improving the safety of autonomous driving systems (ADS) is crucial for the deployment of highly automated vehicles, especially in safety-critical events. To address the rarity issue, adversarial scenario generation methods are developed, in which behaviors of traffic participants are manipulated to induce safety-critical events. However, existing methods still face two limitations. First, identification of the adversarial participant directly impacts the effectiveness of the generation. However, the complexity of real-world scenarios, with numerous participants and diverse behaviors, makes identification challenging. Second, the potential of generated safety-critical scenarios to continuously improve ADS performance remains underexplored. To address these issues, we propose LLM-attacker: a closed-loop adversarial scenario generation framework leveraging large language models (LLMs). Specifically, multiple LLM agents are designed and coordinated to identify optimal attackers. Then, the trajectories of the attackers are optimized to generate adversarial scenarios. These scenarios are iteratively refined based on the performance of ADS, forming a feedback loop to improve ADS. Experimental results show that LLM-attacker can create more dangerous scenarios than other methods, and the ADS trained with it achieves a collision rate half that of training with normal scenarios. This indicates the ability of LLM-attacker to test and enhance the safety and robustness of ADS. Video demonstrations are provided at: https://drive.google.com/file/d/1Zv4V3iG7825oyiKbUwS2Y-rR0DQIE1ZA/view.

Tian Ye, Xiaolong Wang, James Davidson et al.

Humans have a remarkable ability to use physical commonsense and predict the effect of collisions. But do they understand the underlying factors? Can they predict if the underlying factors have changed? Interestingly, in most cases humans can predict the effects of similar collisions with different conditions such as changes in mass, friction, etc. It is postulated this is primarily because we learn to model physics with meaningful latent variables. This does not imply we can estimate the precise values of these meaningful variables (estimate exact values of mass or friction). Inspired by this observation, we propose an interpretable intuitive physics model where specific dimensions in the bottleneck layers correspond to different physical properties. In order to demonstrate that our system models these underlying physical properties, we train our model on collisions of different shapes (cube, cone, cylinder, spheres etc.) and test on collisions of unseen combinations of shapes. Furthermore, we demonstrate our model generalizes well even when similar scenes are simulated with different underlying properties.

Ye Tian, Long Wang

Classic models on opinion dynamics usually focus on a group of agents forming their opinions interactively over single issue. Yet generally consensus can not be achieved over single issue when agents are not completely open to interpersonal influence. In this paper, opinion consensus in social networks with stubborn agents is considered over issue sequences. The social network with stubborn agents is described by the Friedkin-Johnsen (F-J) model where agents are stubborn to their initial opinions. Firstly, we propose some sufficient and necessary conditions both in terms of network topology and system matrix for convergence of the F-J model over single issue. Secondly, opinion consensus of the F-J model is investigated over issue sequences. Our analysis establishes connections between the interpersonal influence network and the network describing the relationship of agents' initial opinions for successive issues. Taking advantage of these connections, we derive the sufficient and necessary condition for the F-J model to achieve opinion consensus over issue sequences. Finally, we consider a more general scenario where each agent has bounded confidence in forming its initial opinion for every issue. By analyzing the evolution of agents' ultimate opinions for each issue over issue sequences, we prove that the connectivity of the state-dependent network is preserved in this setting. Then the conditions for agents to achieve opinion consensus over issue sequences are proposed. Simulation examples are provided to illustrate the effectiveness of our theoretical results.