This paper introduces Hierarchical Diffusion Policy (HDP), a hierarchical agent for multi-task robotic manipulation. HDP factorises a manipulation policy into a hierarchical structure: a high-level task-planning agent which predicts a distant next-best end-effector pose (NBP), and a low-level goal-conditioned diffusion policy which generates optimal motion trajectories. The factorised policy representation allows HDP to tackle both long-horizon task planning while generating fine-grained low-level actions. To generate context-aware motion trajectories while satisfying robot kinematics constraints, we present a novel kinematics-aware goal-conditioned control agent, Robot Kinematics Diffuser (RK-Diffuser). Specifically, RK-Diffuser learns to generate both the end-effector pose and joint position trajectories, and distill the accurate but kinematics-unaware end-effector pose diffuser to the kinematics-aware but less accurate joint position diffuser via differentiable kinematics. Empirically, we show that HDP achieves a significantly higher success rate than the state-of-the-art methods in both simulation and real-world. HDP combines the best of both worlds by chaining a high-level NBP agent with a low-level learned controller. At the high level, HDP takes the 3D visual observations and language instructions as inputs and predicts a 6-DoF next-best end-effector pose. At the low level, given the high-level 6-DoF end-effector pose action as a goal, HDP casts the control task as a context-aware 6-DoF pose-reaching task. We introduce a novel kinematics-aware low-level agent, RK-Diffuser, a diffusion-based policy that directly generates the motion trajectory via conditional sampling and trajectory inpainting. RK-Diffuser learns both end-effector pose and robot joint position diffusion, and distills the accurate but kinematics-unaware end-effector pose trajectory into the joint position trajectory via differentiable robot kinematics. RK-Diffuser achieves accurate trajectory generation and maximum control flexibility, while avoiding violating the robot kinematic constraints. In our experiments, we empirically analyse HDP on a wide range of challenging manipulation tasks in RLBench. We show that (1) RK-Diffuser generally achieves a higher success rate on goal-conditioned motion generation. (2) The proposed hierarchical agent, HDP, outperforms the flat baseline agents and other hierarchical variants. (3) HDP can be directly trained on a real robot with only 20 demonstrations on a challenging oven opening task with a high success rate.This paper introduces Hierarchical Diffusion Policy (HDP), a hierarchical agent for multi-task robotic manipulation. HDP factorises a manipulation policy into a hierarchical structure: a high-level task-planning agent which predicts a distant next-best end-effector pose (NBP), and a low-level goal-conditioned diffusion policy which generates optimal motion trajectories. The factorised policy representation allows HDP to tackle both long-horizon task planning while generating fine-grained low-level actions. To generate context-aware motion trajectories while satisfying robot kinematics constraints, we present a novel kinematics-aware goal-conditioned control agent, Robot Kinematics Diffuser (RK-Diffuser). Specifically, RK-Diffuser learns to generate both the end-effector pose and joint position trajectories, and distill the accurate but kinematics-unaware end-effector pose diffuser to the kinematics-aware but less accurate joint position diffuser via differentiable kinematics. Empirically, we show that HDP achieves a significantly higher success rate than the state-of-the-art methods in both simulation and real-world. HDP combines the best of both worlds by chaining a high-level NBP agent with a low-level learned controller. At the high level, HDP takes the 3D visual observations and language instructions as inputs and predicts a 6-DoF next-best end-effector pose. At the low level, given the high-level 6-DoF end-effector pose action as a goal, HDP casts the control task as a context-aware 6-DoF pose-reaching task. We introduce a novel kinematics-aware low-level agent, RK-Diffuser, a diffusion-based policy that directly generates the motion trajectory via conditional sampling and trajectory inpainting. RK-Diffuser learns both end-effector pose and robot joint position diffusion, and distills the accurate but kinematics-unaware end-effector pose trajectory into the joint position trajectory via differentiable robot kinematics. RK-Diffuser achieves accurate trajectory generation and maximum control flexibility, while avoiding violating the robot kinematic constraints. In our experiments, we empirically analyse HDP on a wide range of challenging manipulation tasks in RLBench. We show that (1) RK-Diffuser generally achieves a higher success rate on goal-conditioned motion generation. (2) The proposed hierarchical agent, HDP, outperforms the flat baseline agents and other hierarchical variants. (3) HDP can be directly trained on a real robot with only 20 demonstrations on a challenging oven opening task with a high success rate.