StarCraft agent Pluto trained purely by self-play, no imitation learning, authors confirm

Pluto's authors confirm the StarCraft agent used no imitation learning from human data and was trained purely via self-play, unlike AlphaStar which relied heavily on imitation, making it a new benchmark for StarCraft agents.

2026-10-08 ~ 2026-10-09 · 3 related posts

• Pluto reaches StarCraft II strength with pure self-play, no human imitation — unlike AlphaStar — TimDarcet · 2026-10-08
• Why Pluto stands out among StarCraft agents: pure self-play, no learning from humans — hansonwng · 2026-10-09
• Pluto dev confirms the StarCraft agent used zero imitation learning from humans — TimDarcet · 2026-10-09