Add class GazeboAgent to implement the common properties and functions for the agent of gazebo-based SocialRobot enviroments
Changes:
Add class GazeboAgent, simplify the codes for playground and simple navigation using the new GazeboAgent class;
Remove random action at the start of each episode (randomness in the beginning should be controlled by tasks);
Faster tests by testing multi-task together;
Changed to sin(), cos() to wrap the range of continuous joint to [-pi, pi). Note that this changes the dimension of observation and make the previous training checkpoint fails to run.
Add class GazeboAgent to implement the common properties and functions for the agent of gazebo-based SocialRobot enviroments
Changes: