This video showcases a custom-built robotic platform alongside its differentiable digital twin, developed within Ostrich, a GPU-accelerated differentiable simulation environment.
The demonstration highlights how real-world robotic systems can be synchronized with high-performance simulation to enable large-scale learning, real-time estimation, and advanced control in challenging environments.

Ostrich is a differentiable rigid-body simulator that resolves hard contacts and friction using non-smooth Newton iteration at an integration step of 0.1 s. By differentiating through the converged residual via the implicit function theorem, the simulator computes adjoints with O(1) memory per timestep.

Key highlights:
🔹 30x faster forward and backward passes
🔹 Substantially more stable gradients
🔹 O(1) memory complexity per timestep
🔹 Up to 2,000x more parallel simulation worlds on the same GPU compared to MuJoCo
🔹 Enables large-scale learning, real-time estimation, and advanced control
About XSCAVE
XSCAVE is a Horizon Europe-funded project that aims to revolutionize the autonomy of heavy mobile machines, including excavators, forestry forwarders, and urban logistics vehicles, through AI-driven terrain-adaptive control systems. The project focuses on improving operational efficiency, safety, and environmental sustainability while addressing labor shortages in sectors that rely on heavy machinery.