Why imitation learning still tops RL on benchmarks

Debate is intensifying over whether imitation learning really needs RL on top to handle distribution shift. Posters note that public benchmarks are still often led by imitation-learning agents such as TFv5/6 and DrivoR, raising questions about whether RL’s claimed robustness advantage is overstated.

2026-07-15 ~ 2026-07-15 · 2 related posts

• Imitation Learning vs RL Distribution Shift Debate — ftm_guney · 2026-07-15
• Does Imitation Learning Really Outperform RL? — srchvrs · 2026-07-15