We use the concept of memory of motion to warm-start trajectory optimization. In this work we use TrajOpt (http://rll.berkeley.edu/trajopt/doc/sphinx_build/html/) as the trajectory optimizer. A set of problems are solved online using TrajOpt, and we use function approximators to predict the initial guesses online for warm-starting TrajOpt.

Install TrajOpt:

see http://rll.berkeley.edu/trajopt/doc/sphinx_build/html/

Install pbdlib-python for BGMR:

see https://gitlab.idiap.ch/rli/pbdlib-python

Install GPy:

see https://github.com/SheffieldML/GPy

• Teguh Santoso Lembono - Idiap Research Institute, Martigny, Switzerland Ecole Polytechnique Federale Lausanne teguh.lembono@idiap.ch