Seeing Machines Limited (AIM: SEE) has published a technical paper demonstrating that robots and other Physical AI systems can achieve real-time 3D human understanding directly on embedded hardware. Announced on September 30, 2026, the paper details the company's Human Mesh Recovery technology, a core capability of its Human-Centered Physical AI Platform launched in August 2026. The research aims to support safer and more natural human-machine interaction in settings where people and machines work in close proximity.
Real-Time 3D Human Understanding
The technical paper focuses on Seeing Machines' Human Mesh Recovery (HMR) technology. Using a single camera view, HMR reconstructs a detailed 3D representation of a person so that a machine can understand their location, posture, and movement in real time. This capability is designed for factories, hospitals, warehouses, and other dynamic environments where robots and people need to work safely and effectively together, requiring machines to interpret human motion accurately and respond without delay.
Overcoming Edge Computing Constraints
Physical AI systems operating around people must do more than detect a person's presence. To respond safely and effectively as interactions unfold, robots need to understand a person's position, posture, and movement in real time without relying on delayed remote processing. Until now, high-fidelity 3D human reconstruction has typically required significant computing resources, making real-time deployment on embedded devices challenging for many applications.
Local Processing on NVIDIA Jetson Thor
The company states that HMR technology addresses this constraint by delivering detailed 3D human understanding at the edge. Processing information locally allows robots to respond to a person's position and movement in real time without relying on a remote cloud connection. The technical paper demonstrates a compelling balance of accuracy, inference speed, and model efficiency when deployed on NVIDIA Jetson Thor hardware, removing the latency and connectivity risks associated with cloud-dependent systems.
Camera Flexibility and Deployment
The HMR model supports both standard RGB cameras and RGB-D cameras without increasing its parameter count. This flexibility makes the technology more accessible for different robotic systems, form factors, and industrial settings where hardware options vary. The results indicate that high-fidelity 3D human reconstruction can be performed in real time on embedded hardware previously considered unsuitable for this level of scene understanding.
Executive Perspective
Paul McGlone, CEO of Seeing Machines, said that robots need to understand more than whether someone is present to work safely and effectively alongside people. He described the HMR technology as delivering a compelling combination of accuracy, latency, and model size on embedded hardware, bringing detailed 3D human understanding into real-world environments. McGlone added that the research demonstrates how Seeing Machines' 25 years of human-sensing expertise can support the next generation of Physical AI.
A Broader Physical AI Platform
HMR is a core capability within Seeing Machines' broader Human-Centered Physical AI Platform. The platform combines the company's expertise in computer vision, human behavior, and safety-critical AI to enable machines to interpret people and their environments in three dimensions. This foundation supports expansion from automotive, commercial fleet, aviation, rail, and off-road applications into robotics and industrial automation where safe collaboration is essential.
Seeing Machines brings more than 25 years of experience developing and deploying human-centered AI solutions across automotive, commercial fleet, aviation, rail, and off-road applications. The new technical paper positions the company's HMR technology as a practical step toward safer robots and industrial automation. By enabling real-time 3D human understanding on embedded hardware, Seeing Machines is expanding its established safety and human-sensing foundation into the next generation of Physical AI and more intuitive robotic systems.