Post by NTU MARS Lab

Excited to share that our paper "EM-Fall: Embodied mmWave Sensing for Day-and-Night Fall Detection on Humanoid Robots" has been accepted to 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)! 🎉🤖 Congratulations to Yanshuo Lu and the team! 👏 Falls are one of the most serious safety risks for older adults living at home. Yet practical in-home fall detection remains difficult: wearable devices depend on long-term user compliance, fixed sensing installations have limited spatial coverage, and camera-based systems can suffer from privacy concerns, poor lighting, and occlusion. 💡 Core Idea: EM-Fall explores an embodied alternative. We deploy millimeter-wave sensing on a mobile humanoid robot, allowing the robot to actively follow users, expand detection area, and maintain target observability across rooms, under occlusion, and in low-visibility conditions. ⚙️ Core technical pipeline: - 📡 mmWave point-cloud clustering and EKF tracking - 🤖 people-as-planner active following for continuous monitoring - 🧹 human-centered perception for non-human target filtering and ghost target suppression - ⏱️ lightweight temporal modeling over pre-fall, drop, and post-fall phases 📊 Results from our real-world evaluation: - 🤖 Platform: Unitree G1 humanoid robot with a TI IWR6843-ISK radar - 🏠 Dataset: normal lighting, low visibility, and occlusion, including fall-like daily activities and pet-like motion interference dataset collected from four volunteers across eight real indoor environments - 🎯 Detection: 100% TPR across all three environmental settings. 100.0% F1-score in normal lighting, 97.1% F1-score in low visibility, and 94.9% F1-score under occlusion - 👀 Cross-modality comparison: EM-Fall remains available across all eight scenes, while vision and LiDAR baselines fail in some low-visibility or occlusion cases From fixed, passive sensing to mobile, privacy-preserving, robot-assisted safety monitoring for real homes. 🚀 🔗 Project: https://lnkd.in/gdVRewzU 📄 Paper: https://lnkd.in/gQFhnP4P #EmbodiedAI #IROS2026 #Robotics #HumanoidRobots #mmWave #FallDetection #ElderlyCare #HomeRobotics #RobotPerception #PhysicalAI #NTU #MARSLab

Video Content