HomeFlow: A Data Flywheel for Smart Home Agent Training with Verifiable Simulation
For smart home agent training, HomeFlow provides a scalable method to generate high-quality, verifiable training data, addressing the bottleneck of data scarcity in physical-world control tasks.
HomeFlow introduces a verifiable data flywheel for training smart home agents, achieving 87.03% task success rate with an 8B model, surpassing GPT-5.5 by 1.23 percentage points.
Large language model agents are moving beyond text-only interaction toward physical-world control, with smart homes as a representative domain. Real domestic interaction requires understanding ambiguous intents, operating in dynamic environments, and performing multi-turn reasoning. However, existing methods struggle to generate high-quality training data for smart home agents. We propose HomeFlow, a verifiable data flywheel for this domain. HomeFlow uses HomeEnv as a unified simulation environment and HomeMaker to procedurally generate diverse home settings. Subsequently, Blueprint compiles open-ended user intents into executable state-based success conditions, while MCTS-Flow synthesizes diverse, verifiable multi-turn trajectories through environment-guided tree search. We then optimize the agents via supervised fine-tuning and step-wise RLVE, which facilitates iterative improvement through authentic physical feedback. We further construct SmartHome-Bench to evaluate the agent across various smart home tasks. On this benchmark, HomeFlow-RL-4B and HomeFlow-RL-8B achieve task success rates of 84.60% and 87.03%. It is worth noting that HomeFlow-RL-8B even surpasses the leading GPT-5.5 by 1.23 percentage points.