2024 was the year Physical AI moved from concept to reality. Whether in robotics, world models, materials discovery, physical simulations, smart homes, or industrial automation, the potential of modern AI foundation models to solve critical real-world problems became clear and entered the mainstream. It was also a transformative year for Archetype AI. Starting with our operational launch in January to pursue the Physical AI vision, we built the initial version of Newton — the first general-purpose physical AI foundation model for the physical world. We initiated collaborations and completed large-scale projects for our customers, shared our vision at TED AI and conferences worldwide, released our first research paper, and grew awareness about the importance of Physical AI. We are proud to be at the forefront of this movement. Let's look at some of the key moments from Archetype AI this year.
Beyond Chatbots: The Untapped Potential of Sensors
At the start of the year, Archetype AI closed a seed round led by Venrock and joined by Hitachi Ventures and the Amazon Industrial Innovation Fund.
The investment enabled us to advance the development of a foundation model for the physical world that we call Newton, moving from concept to reality in our mission to use AI to solve real-world problems across diverse customer use cases and industries.
The timing couldn't have been better. The proliferation of sensors from the IoT era, massive adoption of cloud computing infrastructure, and the shift from machine learning to generative AI architectures provided the foundation for Newton, our large behavior model. It is designed to perceive and reason about the physical world in real time by fusing multimodal sensor data and natural language.
Read on:
- Introducing Archetype AI – Understand the Real World, in Real Time
- Trillion Sensor Economy: How Physical AI Unlocks Real World Data
- WIRED: This AI Startup Wants You to Talk to Houses, Cars, and Factories
- This Week in Startups: Pioneering Physical AI with Archetype AI’s Ivan Poupyrev
An AI Model that Learns Physics from Raw Data
Our research paper, "A Phenomenological AI Foundation Model for Physical Signals," showed Newton's ability to understand and predict real world behaviors without being explicitly trained with underlying principles, i.e., laws of physics.
The research demonstrated how Newton can forecast complex real-world processes it has never encountered before, which unlocks applications of foundation models in a variety of industries and use cases. From predicting power grid demand to understanding oil temperature variations in electrical transformers, our model demonstrated an unprecedented ability to generalize across different physical systems.
Read on:
Can AI Learn Physics from Sensor Data?
Unveiling the Future of Physical AI at TEDAI San Francisco
Venture Beat: Archetype AI’s Newton Model Learns Physics From Raw Data — Without Any Help From Humans
Solving Real World Problems with Physical AI
We demonstrated Newton's potential to solve critical real-world challenges by putting it to work. Our partner Soarchain integrated Newton into their decentralized connectivity platform to create an intelligent road safety system that processes real-time camera data and interprets complex road situations.
With Khasm Labs, we ran Newton on a single edge GPU to process real time traffic data in an intersection and urban data streamed from drones over a 5G network. Our partnerships with Infineon and several Fortune Global 500 customers—spanning automotive, consumer electronics, and construction sectors—validated our vision of using Physical AI to solve real world problems. We were also thrilled to be selected for the AWS Gen AI accelerator to speed up development of real-world applications with Newton.
Read on:
How Soarchain Is Making Roads Safer With Physical AI
Bringing AI to Sensor Data: Newton On-Prem and Real-Time
Infineon to Pilot New AI Developer Model by Archetype AI to Anhance AI Sensor Solution Innovation
Washington Post: Startups Shaping the Future of Generative AI
AI for Physical Awareness
AI working in the physical world cannot rely just on one kind of sensor. A factory, vehicle, or smart home cannot run just on a camera, but needs hundreds of different sensors working together to capture critical processes and events happening in real time. Working in collaboration with Infineon, we demonstrated how Newton could integrate high-level contextual information with real-time data from simple sensors — just a microphone and radar — to understand and describe complex real-world events.
By fusing basic sensor data with context, we demonstrate that AI can achieve human-like understanding of fluid physical situations. This marks a step toward a future where Physical AI will seamlessly understand and respond to complex real-world events—much like humans do. Whether it's improving home safety, enhancing manufacturing efficiency, or designing autonomous vehicles, the ability to connect different types of sensor data opens up endless possibilities.
Read on:
How AI Can Make Sense of the Real World
As we head toward 2025, we're looking forward to continuing to push the boundaries of what's possible with Physical AI. We are thankful to everyone who supported us on our journey this year, shared excitement, advice, and insights. We thank you all! From all of us at Archetype AI, we wish you a great holiday season and a fantastic new year filled with discovery!