Generating Trajectories

Once you have created an ErgodicManager and a TrajectoryManager, you can generate a trajectory.

xd, ud = pto_trajectory(em, tm)

There are a number of optional keyword arguments. Here they are with their default arguments:

verbose::Bool = true
logging::Bool = false
max_iters::Int = 100
es_crit::Float64 = 0.003
dd_crit::Float64 = 1e-6

When the verbose tag is set to true, progress is presented at each descent iteration. You can turn this to false if you are running trajectory generation as an inner component of a larger algorithm.

The logging tag saves a file temp.csv that contains a copy of each trajectory (just the states, not the actions) at each descent iteration. This is mostly there for gif generation (see Visuals).

The max_iters tag is the maximum number of descent iterations allowed.

The es_crit and dd_crit tags are termination conditions based on the ergodic score and directional derivative, respectively. A lot of research suggests using a directional derivative criterion.

An example using some of these tags is shown below:

pto_trajectory(em, tm, max_iters=1000, dd_crit=1e-4, verbose=false)