Courtesy: Qulacomm
| What you should know:
● Dense urban traffic and highway driving can be complex and often dangerous for road users, but crash avoidance technologies such as ADAS can reduce road incidents. 
● Traditional, rule-based planning methods for controlling ADAS functionality can’t scale to include enough potential scenarios. ● The Snapdragon Ride platform employs an AI planner to learn and adapt in real-time as well as a traditional planner as a safety guardrail and verifier. |
The dense urban traffic at crowded intersections with vehicles, two-wheelers, and pedestrians and highly congested arterial roadways, can be complex and often dangerous for road users. Approximately 1.19 million people died in traffic crashes in 2023. In the U.S., 59% of these road fatalities occurred in urban areas, and 73% were at intersections.
Crash avoidance technologies such as advanced driver assistance systems (ADAS) can reduce road incidents, helping to save lives in these complicated scenarios. For example, automatic emergency braking has been shown to reduce front-to-rear crashes by 50% and pedestrian crashes by 27%.
Achieving these results across cities, countries, and driving styles is no small task. Traditional, rule-based planning methods for controlling ADAS functionality often struggle to negotiate and adapt to real-time sensor data in dense urban driving scenarios. These human-defined, logic-based planners rely on pre-specified rules, which can’t scale to include enough potential scenarios for the planner to react appropriately in any given traffic situation.
AI planner
Introducing an AI-based planner into the system can help to handle the massive amount of input coming into a vehicle as it travels through highly variable and dynamic urban environments. Capable of running large language models while simultaneously processing input from multiple perception systems, an AI planner uses a data-driven approach to learn and adapt in real-time.
Because it is a decision-based transformer, an AI planner understands what information is contextually relevant to the scenario, so the driver assistance system can act upon it quickly and effectively. This ability to quickly and holistically process data allows the planner to solve complex urban traffic problems and achieve a more accurate and human-like driving experience.
Best of both with Snapdragon Ride
To provide a human-like experience, the Snapdragon Ride platform employs a hybrid architecture that blends both types of planning. The AI planner is a fully data-driven, transformer-based model, while the traditional planner serves as a safety guardrail and verifier. The models co-exist on the same heterogeneous system-on-a-chip (SoC), running on separate blocks, so there is no computational interference. The AI planner benefits from AI acceleration in the neural processing unit (NPU) while traditional planners run on the central processing unit (CPU).
Validated in both simulations and real-world scenarios, the AI planner has demonstrated its ability to solve complex traffic scenarios, including unprotected turns, navigating roundabouts, and handling dense traffic merges.
Incorporating both traditional and AI planning gives automakers a robust solution for tackling the challenges posed by dense urban environments, allowing them to fine-tune and customize ADAS features to meet unique market needs. The move toward AI planning will help them to create a more human-like driving experience, potentially revolutionizing urban traffic management.
