#United States #Pleasanton #Atomathic #AISIR for Radar #AI Signal Intelligence

Atomathic Unveils AISIR for Radar Technology

On December 15, 2025, Atomathic, previously known as Neural Propulsion Systems, revealed its groundbreaking technology, AISIR for Radar™, designed to elevate radar perception where conventional systems tend to falter. Specializing in physical AI-sensing technology, the company promises enhanced reliability for advanced driver-assistance systems (ADAS) and autonomous driving applications. Traditional radar systems often struggle in cluttered environments, leading to phantom detections and unreliable object recognition. At the core of this issue lies a dual-system architecture that merges rapid response methods with physics-based reasoning.

Understanding AISIR for Radar™

AISIR, short for AI Signal Intelligence Reasoning, utilizes a unique dual-system approach to achieve reliable radar perception in high dynamic range (HDR) settings. Its innovative architecture comprises two interconnected systems:

1. AIDAR (Fast Response Method): This component quickly processes raw radar data, creating a compact interpretation from the chaotic input, emphasizing signal clarity amidst clutter.

2. AISIR (Reasoned Response Method): This layer focuses on deeper analysis, employing physics-informed generative reasoning to contextualize the scattered data over time. It adeptly discriminates between valid signals and noise, resulting in more stable detection.

Bridging the Reliability Gap

Atomathic's introduction of AISIR addresses a persistent reliability dilemma within the automotive industry. Although radar is critical for operating in adverse conditions like fog and rain—where cameras and LiDAR face limitations—existing systems falter in HDR, particularly in detecting vulnerable road users (VRUs). This performance gap often leads manufacturers to underutilize radar, illustrated by Tesla's recent decisions to remove radar sensors from its Full Self-Driving suite in favor of consistency. However, AISIR's physics-based method reframes envisaging radar's potential, significantly improving its accuracy and reliability.

Insights from the White Paper

Accompanying the launch is a white paper titled "Physical AI Reasoning for Stable Radar Perception Closing the Reliability Gap". The publication digs into the technical nuances of conventional radar limitations and delineates how the dual-system architecture uniquely counters these issues. Central to its findings are:

• - Structured Sparse Reconstruction: Solving the challenge of frequent radar reflections overpowering a limited number of antennas in densely packed environments.
• - Dual-System Stability: Evidence showcasing how the real-time processing of AIDAR alongside the generative reasoning of AISIR effectively reduces detection errors like 'ghosts’ and supports continuous tracking even under challenging conditions.
• - Successful HDR Stress Tests: Particularly impressive is the demonstration where AISIR successfully tracks a pedestrian next to a large truck, a scenario known to confound traditional systems. In this test, AISIR maintained consistent detection and reduced object flicker, demonstrating its advanced cognitive capabilities.

Industry Impact

The push towards reasoning-centric radar processing is underscored by recent research from tech giants like NVIDIA and Waymo, validating the necessity of physics-grounded frameworks in the autonomous driving landscape. As per industry experts, the advent of radar that matches LiDAR efficacy in safety-critical situations through advanced reasoning processes could shape future automotive designs.

Dr. Behrooz Rezvani, Founder and CEO of Atomathic, stated, "With AISIR, we aim to dismantle the long-standing hurdles in radar reliability and create a platform that empowers autonomous systems to perform efficiently in the real world."

This significant technological advancement could not only transform the perceptions in autonomous vehicle design but may also position Atomathic at the forefront of the burgeoning market for reliable sensing technologies.