This project studies lifelong smart home agents that must keep working as user routines, device states, home layouts, and environment dynamics change over time. The work centers on DomusMind, a benchmark for evaluating whether IoT agents can adapt under drift rather than only perform well in static settings.

Research Focus

• Lifelong learning and adaptation for smart home agents
• Agent robustness under behavioral, environmental, and device-state drift
• Evaluation protocols for long-horizon IoT decision-making
• Benchmarks for trustworthy and practical household AI agents

DomusMind Benchmark

DomusMind evaluates smart home agents in scenarios where the distribution of tasks and context changes over time. This makes it possible to test how well agents retain prior capabilities, adapt to new routines, and recover from changing household conditions.

Paper Poster