Source description
About the role
About the Role : At Humyn Labs, we are building egocentric video datasets from real-world environments residential, agricultural, manufacturing, and construction to train the next generation of robotic AI models. We believe human-collected, real-world data fundamentally outperforms synthetic or simulation-based data for robotic training. Your job is to prove it. We are looking for a Research Analyst who can rigorously compare Human Labs' datasets against sim/synthetic alternatives, publish compelling research that demonstrates the superiority of real human-collected egocentric data, and help position Human Labs as the definitive source of ground-truth robotics training data. What You'll Own : Robotics Model Performance Research : - Evaluate how robotic models trained on Human Labs' egocentric data perform vs. models trained on synthetic or simulation data - Benchmark across real-world domains : - Residential (household tasks, navigation, object interaction) - Agricultural (field operations, crop handling, terrain variability) - Manufacturing (assembly, quality inspection, tool use) - Construction (site navigation, material handling, safety scenarios) - Track metrics such as task success rate, generalization, robustness, and sim-to-real transfer gap - Continuously publish performance comparisons that highlight real-world data advantages Egocentric Video Dataset Analysis : - Deep-dive into Human Labs' egocentric video datasets understand what makes them uniquely valuable - Analyze dataset characteristics including : - First-person perspective richness and scene diversity - Labeling precision, bounding quality, and annotation consistency - Temporal depth and action continuity - Environmental variability (lighting, motion, noise, terrain) - Compare against publicly available sim datasets (e.g., AI2-THOR, Habitat, Isaac Sim, CARLA) and synthetic alternatives - Identify and articulate what differentiates Human Labs' data quality from other vendors Labeling & Annotation Quality Intelligence : - Develop a structured framework to evaluate and score dataset annotation quality - Focus on what matters for robotics training : - Bounding box precision and consistency - Action and event labeling accuracy - Depth, pose, and spatial annotation quality - Edge case coverage in real-world conditions - Showcase how Human Labs' labeling standards outperform industry benchmarks Research Publishing & Thought Leadership : - Publish research reports, white papers, and blog posts that : - Demonstrate human data superiority over sim/synthetic for robotic training - Highlight performance gaps when models trained on sim data are deployed in the real world- - Position Human Labs as a pioneer in real-world egocentric robotics data - Stay deeply read on : - Robotics learning research (imitation learning, behavior cloning, reinforcement learning from demonstrations) - Egocentric video understanding and first-person AI - Sim-to-real transfer literature - Competing dataset vendors and benchmark ecosystems What We're Looking For : - 1 to 4 years of experience in ML research, robotics data, computer vision, or applied AI - Strong understanding of robotics training pipelines and data requirements - Familiarity with egocentric or first-person video datasets (e.g., Ego4D, EPIC-Kitchens, or similar) - Knowledge of sim/synthetic data platforms (Isaac Sim, AI2-THOR, Habitat, CARLA, or similar) - Experience with dataset evaluation, annotation quality assessment, or benchmarking - Ability to write clear, publishable research for both technical and non-technical audiences - Genuine curiosity about the real-world vs. synthetic data debate in AI Technical Skills : - Python (mandatory) - PyTorch or TensorFlow - Video processing tools (OpenCV, FFmpeg, or similar) - Familiarity with : - Robotics learning frameworks (ROS, LeRobot, or similar) - Annotation and labeling tools (CVAT, Scale AI, Labelbox, or similar) - Evaluation metrics for robotics and video understanding - Experience reading and synthesizing ML research papers - Bonus : hands-on experience with sim environments or robotic datasets Ideal Mindset : - Deeply read on robotics AI, egocentric video, and dataset research - Analytical and detail-oriented able to spot what makes one dataset better than another - Passionate about real-world data and its role in making robots actually work - A strong communicator who can turn data comparisons into compelling research narratives - Excited to build Human Labs' reputation as the gold standard in robotics training data What Success Looks Like in 90 Days : - First research report published comparing Human Labs' egocentric data vs. sim/synthetic alternatives on at least one robotic domain - Benchmarking framework live across 23 robotics or video models - Dataset quality scoring system A
More at KGEN