Obsidian/Zettelkasten/Permanent Notes/Literature Notes/to_process/Runtime Safety Assurance Using Reinforcement Learning-Note.md

# First Pass
**Category:** 
Simulation paper that talks about using a recovery controller in combination 
with a nominal controller that is learned. The boundary to swtich between
the two is a learned boundary.

**Context:** 
Drones

**Correctness:** 
Not very. They do really well in the intro and methodology, but shit hits the fan
when it comes to the results. They don't explain things well, and also don't 
really establish why their learned boundary between nominal and recovery
controllers should be successful.

**Contributions:** 
The biggest contributions these guys make is demonstrating feasibility of their 
reinforcement learned switching from the recovery controller and the nominal 
controller.

**Clarity:** 
Well written until the whole thing came apart at the end.

# Second Pass
I read this a second time but I don't think it's worth it.

Their main contribution is trying to use RL to learn when to switch to a 
recovery controller. 

# Third Pass
**Recreation Notes:**

**Hidden Findings:**

**Weak Points? Strong Points?**