Adaptive AI enables groundbreaking automotive intelligence
Autobrains has introduced its latest breakthrough, designated Liquid AI, which represents a paradigm shift in the field of autonomous driving and Advanced Driver Assistance Systems (ADAS).
Liquid AI addresses challenges that even the best automotive companies struggle to overcome. It combines Autobrains’ signature-based self-learning approach with a modular and adaptive architecture of specialized, scenario-based end-to-end skills.
“While current technologies perform well in handling average conventional driving tasks, they fall short when faced with unexpected real-world driving scenarios that demand greater precision. By using or implementing our Liquid AI, automotive companies can close their AI gaps,” adds Autobrains Founder and CEO, Igal Raichelgauz.
Challenges faced by conventional automotive AI
Edge cases — he infinite variety of conceivable unexpected driving scenarios presents conventional AIs with practically unsolvable tasks. Today’s manually trained black-box systems lack the inherent ability to cover edge cases. Attempts to address this by feeding the systems more labeled images result in a loss of trackability and controllability.
Cost — addressing real-world driving problems by blowing up existing systems with more data, labeling, layers, and computational resources leads to escalating costs and power consumption, resulting in diminishing returns. Achieving a substantial improvement in system accuracy by a factor of 10 requires 10,000 times more computational resources.
Perception-decision disconnect — the missing interplay between perception and decision functions hinders effective and precise decision-making. For the AI to make optimal driving choices, it requires specific information. However, when details are missing or overly complex, precision is compromised, leading to incorrect reactions.
Liquid AI is inspired by the human brain
Autobrains draws inspiration from the human brain, which consists of specialized areas akin to task-specific narrow end-to-end AIs. Just as our brain adapts its architecture based on context — such as light and weather conditions, surroundings, and relevant road users — Liquid AI follows the same approach. Here’s how it works:
Network of specialized narrow AIs — Liquid AI comprises hundreds of thousands of specialized narrow AIs, each designed for specific tasks, making reactions very precise and tailored to the relevant driving scenario. This specialized AI approach enables scalability, ranging from a few tens to hundreds of AIs for ADAS systems, scaling up to thousands for higher levels of automated driving, all the way to hundreds of thousands of AIs for full self-driving automotive applications.
Adaptive architecture — unlike fixed systems, the Liquid AI architecture adapts dynamically to the driving context, activating only relevant modules as necessary. This significantly reduces power consumption and compute requirements, not only resulting in cost savings for the System on Chip (SoC) hardware.
Efficiency and precision — by mimicking the brain’s flexibility, Liquid AI achieves superior performance, cost-effectiveness, and safety.
Liquid AI bridges the gap between conventional AI limitations and the promise of truly intelligent autonomous systems, for example in emerging automotive applications. By mimicking the human brain’s flexibility, Liquid AI achieves superior performance, cost-effectiveness, and more explainability and controllability.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
