Antioch: Bridging the Gap Between AI Simulation and Reality

The Rise of Physical AI: A Distant Promise

The concept of physical AI relies on engineers' ability to program physical agents with the same ease as digital agents. However, this vision remains a distant ideal. Currently, robotics is hindered by a lack of data from physical environments. To address this shortfall, companies are building fictional environments where they can test their machines. Simultaneously, an entire industry is developing around monitoring production lines and workers, aiming to train deep learning models to operate robots.

Simulation as a Promising Alternative

In the face of these challenges, simulation emerges as a promising alternative. By creating detailed virtual replicas of real-world environments, roboticists could obtain the necessary data and workspaces to make significant progress. Antioch, a startup specializing in developing simulation tools for robot developers, has set out to bridge what the industry calls the "sim-to-real gap." This term refers to the challenge of making virtual environments realistic enough for trained robots to operate reliably in the physical world.

A Significant Fundraising Round for Antioch

To achieve this ambitious goal, Antioch recently announced it has raised $8.5 million. This funding round, which values the company at $60 million, was led by venture capital firm A* and Category Ventures, with participation from MaC Venture Capital, Abstract, Box Group, and Icehouse Ventures. Founded in New York in May of last year by Harry Mellsop and four other co-founders, Antioch benefits from the expertise of its founders, some of whom previously worked at Google DeepMind and Meta Reality Labs.

The Importance of Simulation for the Industry

The need for better simulation is crucial for many major autonomy companies. In the autonomous vehicle sector, for example, Waymo uses the Google DeepMind world model to test and evaluate its driving models. In theory, this approach could allow Waymo vehicles to be deployed in new areas with less data collection, thereby reducing the costs associated with scaling autonomous vehicle technology.

A Platform for Small Businesses

Antioch positions itself as a solution for new companies that lack the resources to build physical testing arenas or to drive sensor-equipped cars millions of miles. “The vast majority of the industry does not use simulation, and we now clearly understand that we need to move faster,” says Harry Mellsop.

A Technology Inspired by Cursor

Antioch's leaders compare their product to Cursor, a popular AI-powered software development tool. Antioch allows robot builders to create multiple digital instances of their hardware and connect them to simulated sensors that mimic the data the robot's software would receive in the real world. These environments enable developers to test edge cases, perform reinforcement learning, or generate new training data.

The Challenge of Simulation Fidelity

The main challenge lies in the fidelity of the simulations. It is crucial that the physics in the simulation matches reality to avoid any malfunctions when the model is deployed on a real device. Antioch starts with models built by Nvidia, World Labs, and others, and creates domain-specific libraries to facilitate their use. Working with multiple clients allows Antioch to refine its simulations with a depth of context that a single physical AI company could not match.

A High-Stakes Sector

According to Çağla Kaymaz, a partner at Category Ventures, software engineering and LLMs are beginning to influence physical AI. However, the challenges are different. In the physical world, the stakes are much higher than with mediocre coding tools in the digital realm.

Diverse Applications for Physical AI

Antioch primarily focuses on sensor and perception systems, which are essential for automated cars and trucks, agricultural and construction machinery, or aerial drones. While generalized physical AI capable of replicating human tasks is still far off, Antioch has already attracted the attention of multinationals investing heavily in robotics.

Support from Industry Experts

Adrian Macneil, a leader at the autonomous driving startup Cruise, supports Antioch as an angel investor. He emphasizes the importance of simulation for building safety cases or handling tasks that require high precision. Macneil hopes to see tools emerge similar to those that led to the SaaS revolution to support physical AI.

Towards a Revolution in Physical AI

Harry Mellsop is convinced that the future of physical AI lies in developing autonomous systems primarily in software. Experiments are already underway in this direction, such as that of David Mayo from MIT, who uses Antioch's platform to evaluate LLMs. Mayo tests robots designed by AI models with Antioch's simulator, thus offering a new paradigm for evaluating LLMs.

A Promising Yet Distant Future

Before physical AI becomes a common reality, there is still much work to be done to bridge the gap between digital models and the real world. If this goal is achieved, developers will be able to create a "data flywheel," key to the success of category leaders like Waymo. Companies will then have to choose between developing these tools themselves or acquiring them.