Short Summary: Zoox has developed a comprehensive Fail Operational Response (ZFOR) framework that enables autonomous vehicles to safely continue operation after detecting component faults or failures, rather than simply bringing the vehicle to a stop. This system implements intelligent fault management through graduated response strategies—from velocity constraints to strategic pullover maneuvers—allowing Zoox to maintain service continuity while ensuring passenger safety. ZFOR represents a critical competitive advantage for autonomous ride-hailing services, enabling higher fleet availability and improved passenger experience compared to traditional "fail-safe" approaches. 

Short Summary: Zoox has developed a comprehensive Fail Operational Response (ZFOR) framework that enables autonomous vehicles to safely continue operation after detecting component faults or failures, rather than simply bringing the vehicle to a stop. This system implements intelligent fault management through graduated response strategies—from velocity constraints to strategic pullover maneuvers—allowing Zoox to maintain service continuity while ensuring passenger safety. ZFOR represents a critical competitive advantage for autonomous ride-hailing services, enabling higher fleet availability and improved passenger experience compared to traditional "fail-safe" approaches. 

Short Summary: Abstract: Artificial intelligence (AI) is advancing rapidly and tackling increasingly complex problems with remarkable performance, but ensuring safety remains a major problem. In safety-critical, real-world cyber-physical systems, AI promises a new level of autonomy but is hampered by the lack of safety methods. This talk will present a system theory-inspired perspective on AI safety that interprets the data-generating process and its abstraction by AI systems in a system analysis-driven manner. The resulting paradigm of data control encourages AI engineering to rely on established safety analyses in an interdisciplinary way. Using a top-down approach, a generic foundation for safety assurance is outlined that can be adapted to specific AI systems and applications, enabling future innovation. 

Short Summary:  Human-like driving is critical for autonomous vehicles (AVs) to achieve social acceptance and seamless integration into mixed-traffic environments. However, existing approaches struggle to balance anthropomorphic adaptability with robust safety guarantees – a key deployment bottleneck. To bridge this gap, we propose a game-theoretic framework incorporating multi-agent interaction mechanisms (e.g., cooperative bargaining, competitive risk allocation, and hierarchical decision dominance). Through large-scale simulations and limited field trials, our approach demonstrates: Significant behavioral alignment with collective human driving patterns; Enhanced operational safety while reducing conservatism;  Responsive contextual adaptability to dynamic scenariosThe proposed paradigm effectively resolves fundamental tensions between human-like driving naturalness and safety-critical constraints, advancing practical AV integration.

Short Summary: Conditional imitation learning (CIL) trains deep neural networks, in an end-to-end manner, to mimic human driving. This approach has demonstrated suitable vehicle control when following roads, avoiding obstacles, or taking specific turns at intersections to reach a destination. Unfortunately, performance dramatically decreases when deployed to unseen environments and is inconsistent against varying weather conditions. Most importantly, the current CIL fails to avoid static road blockages.

This talk compares the different flavors of end-to-end autonomous driving with the mediated perception approaches. Then offers effective solutions to aforementioned CIL deficiencies. First, sensor fusion helps overcome the generalization and consistency challenges. Second, efficient Occupancy Grid Mapping (OGM) methods can be utilized to detect partial and full road closures partial, and realize dynamic route planning towards reaching the destination. Third, improving autonomous driving accuracy by learning the relationships between CIL specialist branches via a co-learning matrix generated by gated hyperbolic tangent units (GTUs). The latter solution is from our work awarded latest NCTA best paper honorable mention. The talk also shed light on our publicly available dataset for in-cabin distraction detection: https://heshameraqi.github.io/distraction_detection

Short Summary: The evolution of automated driving architectures has created competitions between model-based and AI-driven motion control systems. This talk explores a solution to combine model-based control techniques (like Linear Parameter-Varying (LPV) and switching controllers) with AI-driven methods such as reinforcement learning (RL) and deep neural networks to enable robust and adaptive vehicle motion control.

The presentation will highlight how this hybrid approach leverages the predictability and interpretability of physics-based controllers alongside the flexibility and adaptability of AI to address complex scenarios such as emergency maneuvers, varying road conditions, and real-time trajectory optimization. This talk will discuss some simulation and experimental results on a real-vehicle platform.

Short Summary: Modern over-actuated vehicle systems depend on precise force coordination to achieve optimal yaw moment control, critical for vehicle stability, safety, and handling. While traditional optimization-based control allocation (CA) methods are effective, they become computationally demanding as actuator complexity grows. This work explores imitation learning as a scalable alternative. We present a comparative study between Behavioral Cloning (BC) and Maximum Entropy Inverse Reinforcement Learning (MaxEnt IRL) for neural network-based CA in over-actuated automotive systems. Both approaches are trained using real world data from a Renault Austral prototype to imitate an optimization-based tire force allocator for yaw control. LSTM architectures are used to capture temporal dependencies. The methods are evaluated across generalization, safety, and computational performance. BC demonstrates low inference latency and strong nominal performance, while IRL achieves similar outcomes, even with reduced training coverage. Under actuator failure, both methods exhibit comparable behavior, consistent with training data characteristics.

These findings suggest that imitation learning could be explored as an alternative to traditional optimization-based control allocation in future high-actuation systems, particularly where computational efficiency and scalability become a concern. However, for the present case, optimization based allocation remains the most reliable and well-performing solution. This study serves as a foundational step toward imitating computationally demanding high-level Model Predictive Control (MPC) strategies using neural nets, enabling safe and efficient deployment in real-time automotive environments.

Short Summary: Advanced control systems and on-board sensing improve vehicle agility and surety, while in this business-oriented presentation, we take a broader view toward mass market deployment. General industry beliefs – along with the corrections that occur afterward – can greatly affect timing. Geopolitical trends are broad, and can include regulatory schemes, trade considerations, and optional programs. Regulatory schemes may mandate certain technologies and lead to deployment surges in some regions but can also discourage short-term deployment if regulatory uncertainty is created. On the other hand, optional programs such as the New Car Assessment Programs (NCAP) – which are different across regions and countries – can be a strong lever toward public education and increasing feature take rates. We therefore return to the problem to be solved, looking to leverage ongoing market momentum to slingshot deployment of life-saving capabilities. The current state and evolution of advanced control systems and on-board sensing provide a basis to ask, “And what else needs to be addressed?”, and offers an inflection point for the integration of ADAS and advanced sensing systems. In changing the focus from communications-as-a-sensor to the modification of an ADAS system to incorporate communications, an incremental step is taken toward mass deployment.