Learning agents · game AI · research

Engineering the first self-taught Tales of Tribute agent

Me presenting at the Somers Science Fair, 2023

via deep RL and abstract action selection.

Tales of Tribute, or ToT, is an online card game that was particularly hard for AI to play. Imperfect information, a large and dynamic action space, and abstract long-term planning each contributed to make the game a particularly tough reasoning challenge for AI to crack.

As a result, agents at the time required human domain knowledge in order to function. While in high school, I engineered the first RL agent for the game, which learned without human knowledge and approximately matched or beat most baseline domain-knowledge agents.

Publication · IEEE Conference on Games 2025

Training a Reinforcement Learning Agent for Tales of Tribute

Sebastian Lashmet & Alexander Dockhorn · 2025

Read the paper on IEEE Xplore ↗

Open PDF ↗ · Download PDF ↓

DOI: 10.1109/CoG64752.2025.11114280

Junior Sciences and Humanities Symposium 2025. This was the New York regional qualifier stage, from which I was selected out of 600+ entrants to go to Nationals in our five-person NY delegation.

Selected research awards

JSHS Nationals 20252nd place in computer science and mathematics
Junior Science and Humanities Symposium (JSHS) 2025, New York region5th place overall
Dr. Robert Pavlica Science Research Award
Downstate Tri-County Science Fair1st place overall
Westchester Science and Engineering Fair 20253rd place in computer science and mathematics
UAlbany Inaugural High School Research Symposium2nd place overall
Somers Science Fair 20233rd place in computer science and mathematics

Email me

lashms@rpi.edu