An implementation of the basic iLQR algorithm focused on code readability (not performance). It includes both the "standard" control-cost Hessian regularization proposed in Jacobson19701 and the value function Hessian regularization from Tassa20122. The line-search implementation also follows Tassa20122 and the regularization schedule is based on values of the line-search parameter3. Currently, no form parallelization is implemented, but it will likely be added at some point in the future.
Footnotes
-
Jacobson, David H. and Mayne, David Q. - Differential Dynamic Programming. - 1970. ↩
-
Tassa, Yuval and Erez, Tom and Todorov, Emanuel - Synthesis and stabilization of complex behaviors through online trajectory optimization. - 2012. ↩ ↩2
-
The idea is not novel, but I do not recall the paper where it is introduced. ↩