The goal of this project is to develop magic tricks for humanoid robots. Robot magic is an interesting benchmark problem to develop novel algorithms for attention and perception, complex motion planning, and human robot interaction. In this project the goal is to implement several well-known card, coin, and scarf tricks. The magic tricks require online generation of complex motion plans to accommodate differences in props and environments. Apart from kinematic constraints, the motions must also satisfy dynamic constraints so that coins remain hidden from the audience and scarfs appear to seemingly appear out of thin air. The motion planner also needs to include additional motions to misdirect the attention of the viewer as to not give the trick away. The motion planner will use reinforcement learning and deep reinforcement learning to implement deterministic and probabilistic motion planning algorithms. Some of the motion plans may also be generated through demonstration by a human. Students will receive a stipend of $13,000 NTD per month and a fee waiver for the lab fee at NTNU. Students must organization, but ERC students will provide some suggestions if so desired.