@sknudstrup presents on the recent changes to Monty’s motor system. He and @tslominski have been doing some major refactoring on this part of Monty, and this video covers a high-level overview of the new shape the motor system has taken. The changes now allow Monty to switch between different policies dynamically. It also makes it much easier to configure more complex policies and customize how Monty interacts with its environment.
Summary Video
Main Video
0:00 Introduction
0:23 We Changed the Motor System
1:47 Why? So we can do this:
2:10 A Real Example
3:18 The Selector Protocol
4:19 Tricky Things Your Selector May Have To Handle
8:23 The Difference between Observations and Perception
10:26 The Difference Between a Selector and a Policy
17:04 An Invitation for the Robotics Community
22:21 Caveat About Surface Policies
24:27 Reframing this as State Machines
26:24 What a State Machine Would Look Like in Python
28:38 Further Discussions