Max Robotics — Global Service Robot Directory + US Certification

About

Jonas Degrave is a robotics and machine learning researcher whose work spans reinforcement learning, differentiable physics, and compliant robotics. He is perhaps best known for pioneering the development of differentiable physics engines for deep learning in robotics — a landmark contribution that challenged the prevailing reliance on black-box optimization methods like evolutionary algorithms by enabling gradient-based controller optimization directly through robotic simulations (176 citations). This work fundamentally shifted how researchers approach robot learning pipelines. Degrave also made significant strides in reinforcement learning with sparse rewards, co-developing Scheduled Auxiliary Control (SAC-X), a paradigm that allows agents to learn complex behaviors from scratch using auxiliary tasks — garnering 155 citations and influencing subsequent work in sample-efficient RL. His earlier research laid important groundwork in compliant and soft robotics, exploring how biologically inspired, soft-bodied quadrupedal robots such as Oncilla and Tigrillo can exploit body dynamics to simplify locomotion control. Across his career, Degrave has demonstrated a rare ability to bridge theoretical machine learning with hands-on hardware experimentation. His cumulative contributions — spanning terrain classification, transfer learning of gaits, and self-supervised image embeddings for control — reflect a researcher deeply committed to making robots that are both intelligent and physically adaptive.