Hello Roboticists, Tomorrow I will be presenting my preliminary exam, 9am Thursday May 4th. You are all welcome to attend either in person or virtually. *Title:* Reinforcement Learning Approaches for Hand Agnostic Manipulation *Date and Time:* Thursday, May 4th, 9:00 AM *Location:* Rogers 226 *Zoom Link: * https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonstate.zoom.us%2Fj%2F93329872744%3Fpwd%3DdytidWNnaUlNZkp1NzdHckt4ajdXdz09&data=05%7C01%7Crobotics-seminar%40engr.oregonstate.edu%7Cb45a6874112445a1790308db4c221ef6%7Cce6d05e13c5e4d6287a84c4a2713c113%7C0%7C0%7C638187481648321408%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=9QsdYLUtcvNeK1zjroYzpfkP%2F7%2BL9Q9WvmRMHhNjI4w%3D&reserved=0 *Password:* 539551 *Abstract:* Robotic grasping and manipulation is a complex domain for control due to a wide array of sensors, hand shapes and sizes being used. In addition, manipulation covers a large number of different tasks, each with different sequences of motions required for completion. This results in controllers often being limited to specific hands, sensing strategies, tasks and even objects. Advances in Transfer Learning and Hierarchical Learning show promise in allowing learned policies to be applied to a broader number of domains than they were initially trained on, as well as allowing policies to leverage lessons from similar tasks to speed up learning on new tasks. I propose applying these advancements to the domain of grasping and manipulation with the end goal of developing manipulation controllers that are robust to changes in hand design and environment and that can be applied to multiple different tasks. Additionally, I explore commonly used sensors in grasping and evaluate their usefulness for learning based approaches. *Advisor:* Dr. Cindy Grimm *Committee Members:* Dr. Ravi Balasubramanian, Dr. Joe Davidson, Dr. Xiaoli Fern, Dr. Mike Bailey Thank you, Nigel Sweson