2026 Intern, On Device Physical AI (Summer)

Samsung Research America Internship•Mountain View, CA

20h•$36 - $63•Onsite

About The Position

You are passionate about building high performance on-device physical AI systems. In this role, you will be researching end-to-end machine learning computation, on-device, real-time vision pipelines with physical AI. The candidate will work with the Samsung robotics team to integrate, design, and innovate efficient on-device physical AI model system and training pipeline for robotics and physical AI (e.g. VLM, VLA), deliver low power, low latency control, and support spatial perception and reasonings for autonomous systems.

Requirements

Currently pursuing a Ph.D. in a relevant technology field at a top ranked university (e.g. Computer Science, Electrical Engineering, other etc.), or Masters with prior relevant industrial experiences
Machine learning computation system or parallel computing for VLM and/or VLA
Prior exposure to robotics simulation experience (Issac Sim, mujoco, etc.)
System architecture (memory, computation) and ML model architecture
On device and/or embedded system programming
Real-time computer vision system, ideally, with 3D real-time vision experiences
In-context learning and/or spatial reasoning research
Good communication skills and good team player

Nice To Haves

Solid understanding of real time vision pipelines, latency tight architectures for autonomous control is a plus
Prior in-context learning and/or spatial reasoning research is a plus
Strong system programming skills, including kernel optimizations, instruction set programming, real-time/embedded programming skills, plus
Worked on physical AI or robotics systems and are familiar with ROS is a plus
Prior publications on the topic is a plus (ICML, ICLR, CVPR, ICCV, ICRA, RSS)

Responsibilities

Conduct research on end-to-end machine learning computation and real-time vision pipelines specifically for physical AI and robotics.
Design, innovate, and test efficient, on-device physical AI model systems and training pipelines (focusing on VLM and VLA architectures).
Implement system architectures that deliver low-power, low-latency control for autonomous systems.
Develop and support adaptive in-scene logic to enhance the capabilities of autonomous robotics.
Perform on-device and embedded system programming, utilizing skills in kernel optimization and instruction set programming.
Work closely with the Samsung robotics team to integrate these new AI models into physical systems (potentially utilizing ROS, but prior ROS experiences not required).