"Test-Time Graph Search for Goal-Conditioned RL" w/ @jwquan_0102@c_voelcker@du_yilun@igilitschenski β accepted to ICML 2026!
We push agents from 0% to >90% success on long-horizon tasks: zero extra training, zero online interactions. π§΅
https://t.co/A7WGHEICmW
[6/6] Next time your robot fails at a long-horizon task, don't rush to retrain it. Try TTGS!
π Paper: https://t.co/A7WGHEICmW
π Site: https://t.co/3ubV6gI1wt
π» Code: https://t.co/K1HlFHtxwL