Tensor utilities#

Genesis World provides utilities for converting between array and tensor formats.

Overview#

Genesis World keeps simulation data in three formats:

  • Quadrants fields: GPU-accelerated physics computation.

  • PyTorch tensors: state access and differentiability.

  • NumPy arrays: data export and visualization.

Conversion helpers#

The helpers below are exposed under gs.utils. They accept Genesis, PyTorch, or array-like inputs and handle any GPU-to-CPU transfer internally.

genesis.utils.misc.tensor_to_array(x: Tensor, dtype: type[numpy.generic] | None = None) → ndarray[source]#
genesis.utils.misc.tensor_to_cpu(x)[source]#
genesis.utils.misc.to_gs_tensor(x, dtype: dtype | None = None)[source]#
genesis.utils.misc.assert_gs_tensor(x)[source]#

See also#

  • Device and platform utilities: the precision-dependent dtype aliases (gs.tc_float, gs.np_float, gs.qd_float, and their integer counterparts), and the globals gs.init() sets.

  • States: state management.