Train An Imaginative Agent
arXiv cs.AI · 20 Oct 2026, 03:59 AEST
A new training framework called Agentic Imagination (Agima) enables agents to anticipate the consequences of their actions before execution. This approach allows agents to simulate actions, explore alternatives, and effectively plan in complex environments.
Key facts
- Agima trains a single omni model to propose actions, generate images of their outcomes, and then reason over these imagined futures to decide what to execute in the real environment.
- Across four environments spanning Maze2D, Sokoban, Pacman, and PushT, Agima generalizes to harder settings unseen during training.
- Trained with SFT alone on two-box Sokoban, Agima solves 40% of four-box puzzles while seeing the real board only once every five moves, compared with an action-only planner that only solves 2%.
- After RL, new behaviors emerge in how Agima uses its imagination: it learns a new unified format that was unseen during the SFT stage, learns to conduct forward thinking only when necessary, and branches more frequently at critical decision points.
Read aloud
A new training framework called Agentic Imagination Agima enables agents to anticipate the consequences of their actions before execution. This approach allows agents to simulate actions, explore alternatives, and effectively plan in complex environments. Agima trains a single omni model to propose actions, generate images of their outcomes, and then reason over these imagined futures to decide what to execute in the real environment. Across four environments spanning Maze2D, Sokoban, Pacman, and PushT, Agima generalizes to harder settings unseen during training. Trained with SFT alone on two-box Sokoban, Agima solves forty percent of four-box puzzles while seeing the real board only once every five moves, compared with an action-only planner that only solves two percent. After RL, new behaviors emerge in how Agima uses its imagination: it learns a new unified format that was unseen during the SFT stage, learns to conduct forward thinking only when necessary, and branches more frequently at critical decision points.
Sources in this cluster
- Train An Imaginative Agent
arXiv cs.AI · 20 Oct 2026, 03:59 AEST