Hi! I'm Diego.
I'm an AI researcher at <stealth>. My goal is to create intelligent systems that can understand their environment and predict the outcomes of their actions to make informed decisions.
The future is exciting!
Publications
Projects
These are some projects I did that were not publication-worthy, but still interesting.
Operational 3D Mesh Synthesis with Geometric and Diffusion Priors
A flexible 3D generative pipeline that uses Score Distillation Sampling to optimize a parametric, production-ready mesh.
An Analysis of Probabilistic Time Series Forecasting with Autoregressive Diffusion Models
A critique of autoregressive diffusion models, which are limited by their deterministic RNN state transition model and single-step objectives.
Report about "Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise"
A review of the paper that claims that diffusion models can revert any non-Gaussian degradation process, including deterministic degradations.
Coupling Grid-based Stochastic Model Predictive Control with Recurrent Neural Networks for Occupancy Grid Prediction
Follow-up work in paper format, improving and scaling up the RNN backbone that predicts traffic motion for trajectory planning in autonomous driving.
Grid-Based Stochastic Model Predictive Control with Occupancy Grids
An RNN to predict the future of discretized traffic environments for trajectory planning with stochastic MPC.
Bio
I’m a founding research scientist at <stealth>, more news coming soon! I earned both my master’s and bachelor’s degrees in electrical enginnering from the Technical University of Munich. During my time at MIT, I was advised by Prof. Vincent Sitzmann, who continued to advise me alongside Prof. Christian Rupprecht at Yellow 3D. My research focuses on building world models as a critical component for (embodied) AI agents to understand and act autonomously in their environment through counterfactual simulation. World models have broad downstream implications for robotics, graphics, and vision.