Difference between revisions of "AgentSmith/Wolfmans Todo List"
Jump to navigation
Jump to search
m |
|||
| Line 10: | Line 10: | ||
* Get the gun to a reasonable standard - add more segmentation | * Get the gun to a reasonable standard - add more segmentation | ||
* Use targeting challenge results to tune gun | * Use targeting challenge results to tune gun | ||
| − | * Weight normalised DC values for unbounded segments in a better manner (such as time-since values etc) | + | * <strike>Weight normalised DC values for unbounded segments in a better manner (such as time-since values etc)</strike> - DONE - ''but performance decreased. Needs investigating'' |
* Factor in distance from the best match when deciding what angle to shoot at based on our K nearest matches. | * Factor in distance from the best match when deciding what angle to shoot at based on our K nearest matches. | ||
* Store dynamic DC weights per bot between matches to give the targeting a head start even if we don't store any of the tree. | * Store dynamic DC weights per bot between matches to give the targeting a head start even if we don't store any of the tree. | ||
* Add in really quick and dirty random movement | * Add in really quick and dirty random movement | ||
* Release into the rumble | * Release into the rumble | ||
| + | * Calculate MEA using [[Precise Prediction]] and [[Wall Smoothing]] | ||
Long Term | Long Term | ||
Revision as of 09:44, 20 March 2013
- Agent Smith Sub-pages:
- AgentSmith - Version History - Challenge Results - Wolfmans Todo List
Short Term
- Get the gun to a reasonable standard - add more segmentation
- Use targeting challenge results to tune gun
Weight normalised DC values for unbounded segments in a better manner (such as time-since values etc)- DONE - but performance decreased. Needs investigating- Factor in distance from the best match when deciding what angle to shoot at based on our K nearest matches.
- Store dynamic DC weights per bot between matches to give the targeting a head start even if we don't store any of the tree.
- Add in really quick and dirty random movement
- Release into the rumble
- Calculate MEA using Precise Prediction and Wall Smoothing
Long Term
- Multiple gun modes chosen dynamically based on match state
- Anti-surfing DC gun mode
- Wave surfing movement
- Genetic algorithms to tune various constants in the bot
- Investigate storing small pre-populated data set for faster DC learning in a fresh match scenario
