[2603.02688] Retrieval-Augmented Robots via Retrieve-Reason-Act

Computer Science > Artificial Intelligence arXiv:2603.02688 (cs) [Submitted on 3 Mar 2026] Title:Retrieval-Augmented Robots via Retrieve-Reason-Act Authors:Izat Temiraliev, Diji Yang, Yi Zhang View a PDF of the paper titled Retrieval-Augmented Robots via Retrieve-Reason-Act, by Izat Temiraliev and 2 other authors View PDF HTML (experimental) Abstract:To achieve general-purpose utility, we argue that robots must evolve from passive executors into active Information Retrieval users. In strictly zero-shot settings where no prior demonstrations exist, robots face a critical information gap, such as the exact sequence required to assemble a complex furniture kit, that cannot be satisfied by internal parametric knowledge (common sense) or past internal memory. While recent robotic works attempt to use search before action, they primarily focus on retrieving past kinematic trajectories (analogous to searching internal memory) or text-based safety rules (searching for constraints). These approaches fail to address the core information need of active task construction: acquiring unseen procedural knowledge from external, unstructured documentation. In this paper, we define the paradigm as Retrieval-Augmented Robotics (RAR), empowering the robot with the information-seeking capability that bridges the gap between visual documentation and physical actuation. We formulate the task execution as an iterative Retrieve-Reason-Act loop: the robot or embodied agent actively retrieves releva...