High Fidelity Dynamics
Hyfydy is software for accurate, high-performance biomechanical simulation. It aims to combine the accuracy of OpenSim with the simulation speed of MuJoCo. Hyfydy is available for Windows, Linux and macOS, as a plugin for SCONE and via a Python API for machine learning applications. Feel free to contact us or request a free trial.
Simulators often use heavily simplified models for muscle and contact dynamics to increase performance. This can lead to inaccurate or low-quality results — especially in machine learning applications, where control policies learn to exploit simulator artifacts.
Hyfydy implements the same high-quality muscle and contact models as OpenSim. Muscle models support tendon elasticity, pennation and fiber damping. Contact models are force-based, with a Coulomb friction cone based on static, dynamic and viscous friction.
Hyfydy has been built from the ground up with maximum performance in mind. It provides an approximate 100x speedup over OpenSim – while using the same muscle and contact models. Simulation speeds are comparable to MuJoCo, even though Hyfydy uses more detailed muscle and contact models, and runs at a much higher internal simulation frequency. In addition, Hyfydy has a low memory footprint, which can be important for large scale optimizations.
Hyfydy uses a novel method for error-controlled integration, which automatically adapts the integration step size according to a user-defined accuracy level. As a result, simulations always remain stable, and always perform optimally within the given accuracy requirement. There is no need for tuning the step size or having to deal with simulations becoming unstable.