Antonio Arbues

I am an independent robotics engineer and researcher interested in e2e learned manipulation, and scalable data acquisition techniques for robotics.

In 2024, I completed my Master's (with distinction) in Robotics, Systems, and Control at ETH Zurich, working with RSL on my thesis about a scalable RL simulator for high-level excavation planning.

Alongside my studies, I spent two years at AMZ Racing, where I led the software team building the autonomy stack of a fully custom self-driving car. I also interned for a semester at Motional where I focused on building a multi-actor e2e model for autonomous driving.

Lastly, I started Seamless, an LLM-based tool that served more than 15.000 students and researchers from the beginning of 2024.

Email  /  Twitter  /  GitHub

Terra is a flexible and abstracted grid world environment for training intelligent agents in the context of earthworks planning. It makes it possible to approach motion and excavation high-level planning as a reinforcement learning problem, providing a multi-GPU JAX-accelerated environment. We achieve SOTA results in the task.

We demonstrate how a careful layering of an LLM in combination with a Model Predictive Control (MPC) formulation allows for accurate and flexible robotic control for contact-rich manipulation tasks via natural language while taking into consideration safety constraints.

I led the software team building the autonomy stack of our custom self-driving car. Most of my time was spent on keeping the team focused on our priorities and debugging software in perception, estimation, control, infrastructure, and hardware integration.

Seamless is a tool that leverages LLMs to help researchers and students find the most relevant papers for their literature review. Shipped in late 2023, it has been used more than 65.000 times by more than 15.000 people from all over the world.