That is, environments featuring the Minitaur environment with a cost associated with the ball/duck falling off the robot and reward determined by forward movement speed. This environment will essentially be like the environment used by David Ha/hardmaru in Evolving Stable Strategiesblog. The lineage for this Minitaur + duck idea goes back further to work by Erwin Coumans.
The version in OpenSafety will split up the balancing the object and locomotion parts of the task into separate cost and reward signals, respectively.
That is, environments featuring the Minitaur environment with a cost associated with the ball/duck falling off the robot and reward determined by forward movement speed. This environment will essentially be like the environment used by David Ha/hardmaru in Evolving Stable Strategies blog. The lineage for this Minitaur + duck idea goes back further to work by Erwin Coumans.
The version in OpenSafety will split up the balancing the object and locomotion parts of the task into separate cost and reward signals, respectively.