A working group session where @vclay and @nleadholm meet with community member @carver to discuss the “Off-Object Observations” work item. The conversation covers information flow and processing in Monty, entry points in the code to start experimenting with on-object vs off-object, and an overview of the Monty architecture and the execution path flow in the sensorimotor loop.
A detailed description of the work item is available here: Use Off-Object Observations
00:00 Processing Off-Object Observations in Monty
00:32 How do you Apply Sensor Data to Update Hypotheses in a Reference Frame?
02:57 Understanding Sensor Location Transformations
14:16 What do we Gain by Processing Off-Object Observations?
23:55 Off Object Observations: Possible Code Entry Points
35:17 A Unit Test to Start With - test_moving_off_object_and_back
37:31 End-to-end Test for Off-Object Information Flow Through Monty
38:47 Handle Failed Jump Exploration
48:18 How Learning Works and Where Off-Object Observations are Filtered
53:18 Off-Object with the Feature Change Filter
56:25 Undo Last Action
59:05 How is Evidence Applied to a Hypothesis?
01:11:43 Stepping Through Monty’s Execution Flow
01:24:08 Exploratory Phase Versus Matching Phase
