Franka Kitchen

## Franka Kitchen Franka Kitchen is a widely-used robot manipulation benchmark dataset that features a Franka Emika Panda 7-DOF robot arm operating in a simulated kitchen environment. The dataset is part of the robosuite/robomimic ecosystem and is one of the most popular benchmarks for evaluating long-horizon imitation learning and offline reinforcement learning algorithms. ### Environment Details - **Robot**: Franka Emika Panda (7-DOF arm with parallel-jaw gripper) - **Environment**: A realistic kitchen simulation with a microwave, kettle, hinge cabinet, sliding cabinet, and burners - **Tasks**: Multi-stage manipulation including opening/closing microwave, opening/closing cabinets, turning on/off burners, operating kettle ### Key Characteristics - **Long-Horizon Tasks**: Tasks require executing multiple sub-skills sequentially - **Sparse Rewards**: Many tasks have sparse reward structure - **Multi-Task**: Supports both individual sub-task learning and full composite task learning - **Demonstrations**: Human-teleoperated demonstrations with multiple trajectories ### Usage in Research The Franka Kitchen benchmark has been used extensively in: - Hierarchical imitation learning (HIL, HIQL) - Offline reinforcement learning (CQL, IQL, TD3-BC) - Behavior cloning and behavioral priors - Goal-conditioned policy learning ### Related Work The kitchen environment was introduced as part of robosuite (arXiv:2009.12293). The dataset has been used in multiple prominent robot learning papers and is available through the robomimic and D4RL ecosystems.