Rafael Possas

Papers

Total Citations

H-Index

About

Rafael Possas is a researcher at the forefront of bridging simulation and reality in robotics, with a primary focus on probabilistic inference, domain randomization, and simulation-based control. His most impactful contribution is the development of **BayesSim**, a groundbreaking framework introduced in his 2019 paper (13 citations) that provides a full Bayesian treatment for simulator parameters. This work fundamentally advances adaptive domain randomization, enabling robots to learn robust policies by probabilistically inferring the most likely simulation parameters that match real-world dynamics. Possas further extended this line of research with **DISCO** (Double Likelihood-free Inference Stochastic Control, 2020), which tackles the challenge of controlling complex physical systems when analytical models are intractable. By combining likelihood-free inference with stochastic control, DISCO allows for more accurate and reliable control strategies before deployment. Though early in his career, Possas’s work is already shaping how researchers think about sim-to-real transfer, offering principled statistical methods to reduce the reality gap. His contributions are particularly valuable for students and engineers working on reinforcement learning, robotics manipulation, and autonomous systems where simulation fidelity is critical for safe and effective real-world performance.