Frozen SAC robot policy, live browser inference and a reproducible CPU benchmark, with source provenance and explicit validation limits.
Dev Kumar
devkumarhf
AI & ML interests
R&D Engineering Consultant — Robotics, Control, Machine Learning & AI. M.Eng., Arts et Métiers.
Reusable robot-learning models, documented datasets and browser-local inference. Featured release: UR7e SAC Line Following — an existing frozen policy with original weights, portable NumPy actor, source provenance and a 600-episode CPU evaluation against a fixed classical controller.
Model: https://huggingface.co/devkumarhf/ur7e-sac-line-following
Live inference: https://huggingface.co/spaces/devkumarhf/UR7e-Policy-Lab
GitHub hosts engineering source and reproduction; Kaggle hosts the measured dataset and experimental notebook; HAL and ORCID provide scientific records and research identity. Existing ICP Lab, Engineering Workbench and the UR7e training-log dataset remain available. Kinematic simulation and physical validation are distinguished.
Open to opportunities in Paris, France; Geneva & Zurich, Switzerland; San Francisco & New York, United States. Open to worldwide freelance projects.
Recent Activity
liked a dataset about 2 hours ago
devkumarhf/ur7e-line-following-training-log liked a model about 2 hours ago
devkumarhf/ur7e-sac-line-following updated a collection about 3 hours ago
UR7e Robot Learning: Model and Live InferenceOrganizations
None yet