Week 6 Competition - CAS

[tawheeler]

For the week 6 competition, students will develop their own collision avoidance systems. The collision avoidance systems will be filled out in provided .jl files, which will then be submitted. The objective is to minimize the number of collisions and the number of alerts (with some weighting between them).

The CAS implements reset() and tick(). They cannot use any Julia libraries besides those used for the class (AA120Q.jl, Distributions, BayesNets, etc.)

The pilot response model will be simplistic and assume that the RA is followed exactly and immediately.

TEACHERS provide:

• the API
• a set of pre-generated human traces from RLES-SISLES
• an example CAS that works
• a pilot response model
• code for running a CAS on the pre-generated human traces
• code for running a CAS on the full batch of human traces from which evaluation metrics are extracted
• server runs 100,000 sims for each student using their CAS to compute a score

STUDENTS provide:

• their own implementation of a CAS by extending our provided CAS .jl file

[tawheeler]

TCAS implemented as of b6fc2f1d. No need for pilot response model (rather, pilots perfectly follow advisory).

[tawheeler]

DemoCAS implemented as of 4667668. Need to collect evaluation metrics and run on server. Need to provide example jl file.

[tawheeler]

Running CAS on full batch (or sampled batch) and print eval stats as of 7a80710

[tawheeler]

We should change the objective function to include the severity of command - ie a climb rate of 1500/60 is not as bad as a climb rate of 4500/60