Tool use is a fundamental capability for robots to operate effectively in human environments. Our research focuses on enabling robotic agents to perform tool-related tasks — such as manipulating a tool (e.g., hammering a nail) or selecting the appropriate tool for a given task — through multiple complementary approaches. We have developed a hybrid method that combine learning-based techniques with motion optimization. By bridging high-level reasoning and low-level control, we aim to build robust and generalizable manipulation capabilities that transfer across different tools and tasks. We are also continuing to explore alternative strategies within the same problem space.