HTTP/2 301 server: GitHub.com content-type: text/html location: https://yjy0625.github.io/projects/globdiff/ access-control-allow-origin: * strict-transport-security: max-age=31556952 expires: Mon, 29 Dec 2025 05:47:49 GMT cache-control: max-age=600 x-proxy-cache: MISS x-github-request-id: 6D49:21D6A4:86733E:96FAFD:695213AC accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 05:37:49 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210040-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1766986669.185300,VS0,VE210 vary: Accept-Encoding x-fastly-request-id: aefee703249f251972aec724b0a26eb359cd6165 content-length: 162 HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 last-modified: Thu, 25 Sep 2025 06:50:57 GMT access-control-allow-origin: * strict-transport-security: max-age=31556952 etag: W/"68d4e651-26bf" expires: Mon, 29 Dec 2025 05:47:49 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: A35C:3ABDEF:843053:94B959:695213AC accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 05:37:49 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210040-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1766986669.414978,VS0,VE213 vary: Accept-Encoding x-fastly-request-id: b166c83fb843608f5806eaff34ae23ab324d9904 content-length: 2802

Rethinking Optimization with Differentiable Simulation
from a Global Perspective

• Rika Antonova^*1
• Jingyun Yang^*1
• Krishna Murthy²
• Jeannette Bohg¹

• Stanford University¹
• MIT²
• ( * Equal Contribution )

Differentiable simulation is a promising toolkit for fast gradient-based policy optimization and system identification. In this work, we study the challenges that differentiable simulation presents when it is not feasible to expect that a single descent reaches a global optimum. We analyze the optimization landscapes of diverse scenarios and find that in dynamic environments with highly deformable objects and fluids, differentiable simulators produce rugged landscapes with useful gradients. We propose a method that combines Bayesian optimization with semi-local leaps to obtain a global search method that can use gradients effectively and maintain robust performance in regions with noisy gradients. We show extensive experiments in simulation, and also validate the method in a real robot setup.

Rethinking Optimization
with Differentiable Simulation
from a Global Perspective

• Rika Antonova^*1
• Jingyun Yang^*1
• Krishna Murthy²
• Jeannette Bohg¹

Stanford University¹ MIT²

Differentiable simulation is a promising toolkit for fast gradient-based policy optimization and system identification. In this work, we study the challenges that differentiable simulation presents when it is not feasible to expect that a single descent reaches a global optimum. We analyze the optimization landscapes of diverse scenarios and find that in dynamic environments with highly deformable objects and fluids, differentiable simulators produce rugged landscapes with useful gradients. We propose a method that combines Bayesian optimization with semi-local leaps to obtain a global search method that can use gradients effectively and maintain robust performance in regions with noisy gradients. We show extensive experiments in simulation, and also validate the method in a real robot setup.