Takeru Oba

Papers

Total Citations

H-Index

About

Takeru Oba is a rising researcher at the forefront of robot motion planning and imitation learning, with a focus on generating feasible, long-horizon behaviors from visual inputs. His work uniquely bridges retrieval-augmented generation and diffusion models for robotics. In his most prominent paper, "READ: Retrieval-Enhanced Asymmetric Diffusion for Motion Planning" (2024, 3 citations), Oba proposes a novel framework that retrieves an initial motion from a database of image-motion pairs and refines it using a diffusion model, enabling efficient and robust planning for novel scenes. He further advances the field with "Cold Diffusion on the Replay Buffer: Learning to Plan from Known Good States" (2023, 2 citations), which explicitly addresses the critical issue of plan feasibility in learning from demonstrations—a challenge often overlooked in powerful imitation techniques. His work on "Future-guided offline imitation learning for long action sequences" (2023, 2 citations) tackles the complexity of generating extended, coherent robot actions by integrating video interpolation and future-trajectory prediction. Though early in his career, Oba’s contributions are already shaping how robots can learn from data and plan reliably, offering practical pathways for deploying learned policies in real-world settings.