EquiBot is a SIM(3)-equivariant diffusion policy for generalizable and data-efficient learning. The method combines SIM(3)-equivariant neural networks with diffusion models to ensure policies are invariant to scale, rotation, and translation, enhancing their applicability to unseen environments while retaining the benefits of diffusion-based policy learning, such as multi-modality and robustness. The method is evaluated on six simulation tasks and six real-world mobile manipulation tasks, showing reduced data requirements and improved generalization to novel scenarios. In real-world experiments, the method generalizes to novel objects and scenes after learning from just 5 minutes of human demonstrations. The method outperforms prior methods in out-of-distribution generalization and data efficiency. The method uses point clouds as input observations and outputs sequences of robot end-effector velocity actions. The method is designed to be robust and generalizable, with a focus on handling a wide range of household manipulation tasks. The method is evaluated in both simulation and real-world experiments, showing superior performance compared to baselines in both settings. The method leverages equivariance to improve data efficiency and generalization, and is designed to handle a variety of tasks involving rigid, articulated, and deformable objects. The method is implemented with a SIM(3)-equivariant diffusion policy architecture, which is trained to be robust and generalizable. The method is evaluated on a variety of tasks, including pushing a chair towards a desk, closing the door of a laundry machine, folding towels, making the bed, and closing a suitcase. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both. The method is designed to handle a wide range of tasks and is capable of generalizing to unseen scenarios with minimal data. The method is evaluated on a variety of tasks, including those involving multi-robot tasks and deformable objects. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both. The method is designed to be robust and generalizable, with a focus on handling a wide range of household manipulation tasks. The method is evaluated on a variety of tasks, including those involving multi-robot tasks and deformable objects. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both.EquiBot is a SIM(3)-equivariant diffusion policy for generalizable and data-efficient learning. The method combines SIM(3)-equivariant neural networks with diffusion models to ensure policies are invariant to scale, rotation, and translation, enhancing their applicability to unseen environments while retaining the benefits of diffusion-based policy learning, such as multi-modality and robustness. The method is evaluated on six simulation tasks and six real-world mobile manipulation tasks, showing reduced data requirements and improved generalization to novel scenarios. In real-world experiments, the method generalizes to novel objects and scenes after learning from just 5 minutes of human demonstrations. The method outperforms prior methods in out-of-distribution generalization and data efficiency. The method uses point clouds as input observations and outputs sequences of robot end-effector velocity actions. The method is designed to be robust and generalizable, with a focus on handling a wide range of household manipulation tasks. The method is evaluated in both simulation and real-world experiments, showing superior performance compared to baselines in both settings. The method leverages equivariance to improve data efficiency and generalization, and is designed to handle a variety of tasks involving rigid, articulated, and deformable objects. The method is implemented with a SIM(3)-equivariant diffusion policy architecture, which is trained to be robust and generalizable. The method is evaluated on a variety of tasks, including pushing a chair towards a desk, closing the door of a laundry machine, folding towels, making the bed, and closing a suitcase. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both. The method is designed to handle a wide range of tasks and is capable of generalizing to unseen scenarios with minimal data. The method is evaluated on a variety of tasks, including those involving multi-robot tasks and deformable objects. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both. The method is designed to be robust and generalizable, with a focus on handling a wide range of household manipulation tasks. The method is evaluated on a variety of tasks, including those involving multi-robot tasks and deformable objects. The method is shown to perform well in both simulation and real-world settings, outperforming baselines in both.