Design Principles#

Each problem class, LQR or iLQR, has a sequential and a parallel solver. Each solver also has a differentiable wrapper with implicit gradients.

Solver type	Module	Entry point	Suited to	Arguments
Sequential	`diffilqrax.lqr`	`solve_lqr()`	CPU	`LQRParams`
Sequential	`diffilqrax.ilqr`	`ilqr_solver()`	CPU	`System, iLQRParams`
Parallel	`diffilqrax.plqr`	`solve_plqr()`	GPU	`LQRParams`
Parallel	`diffilqrax.parallel_ilqr`	`pilqr_solver()`	GPU	`System, iLQRParams`
Differentiable	`diffilqrax.diff_lqr`	`dlqr()`	CPU	`ModelDims, LQRParams`
	`diffilqrax.diff_ilqr`	`dilqr()`	CPU	`System, iLQRParams`
	`diffilqrax.parallel_dlqr`	`pdlqr()`	GPU	`ModelDims, LQRParams`
	`diffilqrax.parallel_dilqr`	`pdilqr()`	GPU	`System, iLQRParams`

The layout of the modules are hierarchically structure outlined in the schematic below: