From Smarter Cars to Safer Cities: Physical AI Solutions that Matter to Society

AI has inspired the imagination of many. Conversations abound about a Jetsonian age of complete automation where humans will have little to do. Others opine that despite the large energy expense of AI, the AI-driven solutions of tomorrow will deliver a net positive impact. 

In a recent conversation, a person suggested that “with AI, we can prevent urban wildfires.” Indeed, AI-driven solutions can rise to address challenges to our quality of life due to the gaps between supply and demand caused by social, economic, and ecological factors.

Some of the factors causing this imbalance include:

• Resource constraints

• Aging populations

• Human capital constraints

• Lack of subject matter experts in critical fields

• Physical security risks

• Supply chain and supply side resiliency

• Negative externalities, such as climate change and environmental pollution

Domain, Data, Data Science

Transforming physical AI ideas into successful solutions necessitates domain knowledge and systemic end-to-end thinking and execution. In the case of urban wildfires, the application of domain knowledge, weather simulation, and a range of AI-driven models based on sensing data can provide insights and inference. The simulations can also provide a window to act and prevent widespread destruction.

However, inference alone is not sufficient. Given the small window, the firefighting actuators must be ready to act on the fires. Actuators include: 

• Supply-side reservoirs of water that are primed and ready

• Fire tenders positioned appropriately with trained human resources ready to act 

• Air-side actuators, such as airplanes and drones that are ready to act. 

For future expeditious action, and given the lack of trained human resources, these actuators would have to be automated. They will also need a policy-based control system that drives closed-loop actions through robotic devices and autonomous firefighting devices.

Introducing the AI-driven Cyber-Physical Stack

Creating such solutions that matter to society necessitates the implementation of a cyber-physical stack comprised of physical and cyber technologies. A full stack starts with deep domain understanding of physical systems and designing the physical systems with appropriate actuators that can be driven using sensing and inference. 

Our conversation must shift from an information technology (IT) stack to a cyber-physical stack. Unlike the IT stack that starts with the user and ends with the database, a full-stack physical AI developer has knowledge of physical and cyber systems.

In this context, I am reminded of the time in 2014 when our daughter announced that she had landed an interview at Tesla. In preparing her for the interview, I mentioned to her that the leader of Tesla has articulated a clear vision of AI-enabled self-driving vehicles. I suggested she step up to the whiteboard in our family room and holistically think through the cyber-physical system—then the Tesla Model S electric vehicle—and draw the cyber-physical stack necessary to achieve the vision.

The Cyber-Physical Stack of an Electric Vehicle

Together we painted the cyber-physical stack with a series of questions that started with the basics of the hardware design, sensing, and actuation, before going into machine-generated data, AI algorithms, and software applications.

A: Sampling of Questions to Understand Physical Design:

1. How does one size the electric motor in the car? 

   a. Estimate the rolling friction, the wind resistance, and which of these two dominate at given speeds.

2. Why did Tesla choose an induction motor? The supply-side battery is 85 KWh. 

3. Given the earlier motor power draw analysis, what is the range of the vehicle?

4. With an 85 KWh battery on the Tesla Model S, at what distances are the 120 KW input power supercharging stations to be located to enable cross-country anxiety-free driving?

5. Imagine a drive from San Francisco to Los Angeles, or San Francisco to Portland, Oregon, on a summer day. 

   a. Given the topography and the climate (for air conditioning load), what would the spacing in miles (km) of the supercharging stations be?

6. Is there potential to build charging stations augmented with solar panels?

   a. Example: What if the 640 km long California aqueduct that flanks Interstate 5 to Los Angeles is covered with solar panels?

7. In the design of the Lithium-Ion battery pack, the thermal management solution to keep the battery temperature within a given temperature range (^oC) at a given location is critical to ensure long battery life.

   a. How would one create and optimize this sub-system for cooling and heating the battery to keep it within a given range using continuous temperature data? 

   b. What are some approaches in thermal management? 

   c. What are the hooks, the knobs, that one would implement to control the thermal management system.

The questions shown above are important to imbue the system and understand the role AI can play in the design and operation of the vehicle. Examples such as AI-driven generative computer-aided design, optimization, and energy-efficient operation help execute seamless and safe journeys.

B: Actuators and sensors to enable future AI-driven vehicles:

The company envisages self-driving vehicles in the future. Here are some pending issues to address:

• In the year 2014, what actuators would the designer have to put into the vehicle such that future algorithms implemented in software can auto-pilot the vehicle?

• What sensing sub-systems have to be pre-built into the vehicle? Indeed, it would be difficult to have millions of cars upgraded in the future if the sensing is not compatible with AI-driven inference.

• An immense amount of data will be collected from the vehicles: from structured (such as temperature in degrees Centigrade from specific locations in a battery cooling system) to unstructured video data. 

  ➜ How would one ensure data discipline to achieve actionable outcomes while reducing energy consumption in transmitting, processing, and storing the data?

• What type of computation will reside on the vehicle, and what amount of machine-generated data will go to the cloud?

• How would historical data mining from thousands of vehicles in the future help in designing infrastructure? 

  ➜ Example: Can the vehicles report anomalies in roadways, and can there be closed-loop action to fix the anomalies?

After our daughter joined Tesla in 2014, she bought a Tesla Model S built with all the physical hooks (the technology package). The vehicle then underwent continuous seamless updates: from pre-production to production versions of various software to include auto-pilot and summon features. The vehicle continues to ply the roads with its original battery pack.

The Road Ahead

The future of successful physical AI is systemic delineation and integration of physical and cyber technologies, and hardware-software co-design. Success will necessitate depth in physical technologies and breadth in cyber technologies. This will beget the rise of T-shaped contributors.

About the Author

Former SVP, Chief Engineer, and Senior Fellow at HP, Chandrakant is a leader in AI, energy-efficient computing, and sustainability. He is an IEEE Fellow, ASME Fellow, member of the NAE, and the Silicon Valley Engineering Hall of Fame. Follow me on LinkedIn or email me at [email protected].

Want to stay ahead of the curve with insights into the newest advancements in Physical AI? Subscribe to Chandrakant’s newsletter at GenAI Works.

🚀 Boost your business with us—advertise where 10M+ AI leaders engage

🌟 Sign up for the first AI Hub in the world.

📲 Our Socials