Thrilled to share some incredible research coming out of Cornell University! Jiaying Fang and her team have introduced a game-changing framework called E-MPC. In human-centered robotics, true autonomy isn't just about a machine finishing its task perfectly. If the user ends up feeling completely passive, disengaged, or overwhelmed, the interaction is not optimal. Instead of treating humans as a simple "backup plan" when the robot gets confused, E-MPC treats user engagement and mental workload as core priorities, predicting exactly when to step in and when to back off. And guess what? They used our Gen3 robotic arm! 🦾 To test the framework, the team ran a real-world bite acquisition study. The data speaks for itself: 9 out of 10 participants preferred this collaborative approach over traditional baselines. It gave them a true sense of agency and satisfaction without adding any extra mental strain. Massive congratulations to Jiaying Fang, Joyce Yang, Zhanxin Wu, Bohan Yang, and Tapomayukh Bhattacharjee on this phenomenal team project. We are so proud to see Kinova hardware supporting this kind of groundbreaking work! 👏 👉 If you are looking for a flexible, ultra-lightweight, and reliable manipulator for your next project, you should definitely check out our page: https://lnkd.in/e_jFpKdE
Task success is not enough for human-centered robots. A robot can complete the task and still leave the user feeling passive, disengaged, or overloaded. Check out our work E-MPC: An Engagement-Aware Human-in-the-loop Framework for Robotic Systems at #RSS2026. E-MPC helps robots jointly reason about task success, user engagement, and user workload. Instead of treating humans only as “backup” when the robot fails, E-MPC reasons about when and how to involve the user throughout the task. E-MPC builds an interaction dynamics model that predicts how the user’s engagement changes over time. The model captures how different interaction choices can shift user engagement closer to or farther from the user’s desired level. E-MPC also models user workload as a constraint. The robot tracks how much user workload it has placed and avoids asking too much. This lets E-MPC balance meaningful involvement with manageable workload, rather than simply maximizing user input. With these models, E-MPC performs model predictive control over both task success and user state. At each step, it chooses whether to act autonomously or ask the user different types of questions, optimizing for task success, engagement regulation, and workload constraints over a short horizon. We evaluate E-MPC in simulation across different user personas and in a real-robot bite acquisition user study. The user study results demonstrate that E-MPC improves engagement tracking, interaction satisfaction, and perceived agency, while maintaining task success and keeping workload manageable. 9/10 participants preferred E-MPC over the baseline. Check out our paper and website! Grateful for my collaborators Joyce Yang, Zhanxin Wu, Bohan Yang, and Tapomayukh Bhattacharjee! Paper: https://lnkd.in/gEHB82wu Website: https://lnkd.in/gYCm4jhU