Presenting "TARAD" at ICRA 2026

I will present our paper “TARAD: Task-Aware Robot Affordance-Centric Diffusion Policy Learned from LLM-Generated Demonstrations” at the IEEE International Conference on Robotics and Automation (ICRA 2026), held June 1–5, 2026 in Vienna, Austria.

TARAD is a framework that uses large language models (LLMs) and vision-language models (VLMs) to perform high-level planning from natural language instructions and to extract affordance information, enabling the training of a multi-task 3D diffusion policy without relying on expert demonstrations or predefined motion primitives. Empirical evaluations in both the RLBench simulation environments and real-world experiments with a UR5e robot show that TARAD combines the precise control of imitation learning with the strong generalization of foundation models.

• Session: TuI1I Interactive Session 1 (Hall C)
• Date & Time: Tuesday, June 2, 2026, 09:00–10:30
• Authors: Site Hu, Takayuki Nagai, Takato Horii (The University of Osaka)

The paper is published in IEEE Robotics and Automation Letters. See the paper highlight page for more details.