Train An Imaginative Agent
We introduce Agentic Imagination (Agima), a training framework that trains an agent to imagine before acting, i.e. to be imaginative.
ProofPaper ↗
Key points
- A seemingly reasonable push in Sokoban can make the puzzle unsolvable.
- To avoid such mistakes, an agent has to anticipate the consequences of its actions before execution, especially when feedback from the real environment is sparse or actions are irreversible.
- This enables the agent to predict future outcomes and perform basic forms of imagination, such as forward thinking and counterfactual branching.
- We then use agentic reinforcement learning (RL) to teach the agent to reason with its imagined futures and use them effectively for planning.
Sources (1)
- [1]Train An Imaginative AgentarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 05:59 PM
We introduce Agentic Imagination (Agima), a training framework that trains an agent to imagine before acting, i.e. to be imaginative.
A seemingly reasonable push in Sokoban can make the puzzle unsolvable.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026UniJev: Separating Accuracy from Calibration in Trimodal System One Decision Models
- Oct 19, 2026TorchNeuron: running HF models on AWS Trainium is easier than ever
- Oct 18, 2026Think Visually Across Space and Time: Grounded Reasoning for Video Segmentation
- Oct 18, 2026CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents
- Oct 18, 2026CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?
- Oct 17, 2026Blackbox Performance Analysis of LLM-based Network Control