Expert Intervention Learning

Learning from Demonstration in Embedded Hardware for the Composition of Micro-Skills in Mobile Robots

“…physical robots for tasks such as road following Pomerleau (1989), trajectory imitation Tan (2015) Vuković et al (2015), or for complete navigation systems Gamal et al (2020), Magnenat and Colas (2021), Caccavale and Finzi (2019) and Spencer et al (2022). As computational platforms, most authors have employed computers for offline training Vuković et al (2015) or processing in real-time Caccavale and Finzi (2019), and in some cases, the trained model has been embedded in microcontrollers for the imitation process, as seen in Tan ( 2015), Vuković et al (2015), Magnenat and Colas (2021) and Spencer et al (2022). In this work, we explore hardware acceleration using an SoC FPGA device to enable online training on a physical robot.…”

“…Physical implementations have also been developed in software, as demonstrated in Vuković et al (2015), where it was implemented on the Kepera II robot, or in Tan (2015), where LfD was applied to the E-Puck robot. In the same way, Magnenat and Colas (2021), Caccavale and Finzi (2019), and Spencer et al (2022) were implemented on the marXbot and Artor mobile platforms using desktops and the last author, implemented the LfD on the NVIDIA Jets device. Table 1 provides a summary of previous works that have implemented LfD in mobile robots, highlighting the learning technique, the mobile robot application, the use of micro-skills, and the computational platform.…”

“…Nevertheless, the training phase was conducted on an external Intel Core i9-8950HK. Furthermore, in Spencer et al (2022), an Expert Intervention Learning (EIL) technique combined with Model Predictive Control was implemented, utilizing an NVIDIA Jetson Nano for a collision avoidance task. However, the proposed technique was trained and tested solely in a single rectangular scenario featuring a straight hallway and sharp corners.…”

Special issue on the 2020 “Robotics: Science and Systems” conference

“…In addition, the authors present a new dataset of verbobject pairs denoting a potential object usage for a variety of objects and usages. The article "Expert Intervention Learning: An online framework for robot learning from explicit and implicit human feedback" by Spencer et al (2021) makes an observation that expert feedback, whether made via intervention or not, provides information about the quality of the current state of the robot, the quality of the action it executes, or both. The authors then utilize this observation to constrain the learner's value function and provide an efficient online learning approach for it.…”