Clinton Enwerem Doctoral Candidate in ECE | Policy Learning for Safe & Robust Poly-Articulated Dexterous Robotic Manipulation

I am an Electrical & Computer Engineering (ECE) Ph.D. candidate at the University of Maryland (UMD), College Park, MD, USA, working with Professor John S. Baras, Distinguished University Professor and Endowed Lockheed Martin Chair in Systems Engineering, and Professor Calin Belta, Brendan Iribe Endowed Professor of ECE and Computer Science. I work on contact-rich poly-articulated (humanoid, serial-chain, and fixed-base bimanual) dexterous grasping and manipulation under uncertainty, drawing on safe and risk-sensitive control, belief-space kinodynamic motion planning, and policy learning to make multifingered grasping and manipulation more reliable when object pose, friction, and contact modes are only partially known. At UMD, I am affiliated with the Institute for Systems Research, the EXplainable and Assured Control for AuTonomy (EXACT) lab, the Systems Engineering and Integration lab, and the Maryland Robotics Center (MRC). I have also received the Dean's Fellowship from UMD's Graduate School and the Microsoft Diversity in Robotics and Autonomy PhD Fellowship, awarded through a collaboration between Microsoft Corporation and MRC.

Prior to resuming doctoral studies at UMD, I was a robotics engineer at Kognitive Robotics, a local robotics engineering startup building turnkey mobile robot platforms for education and research. Before that, I completed a year-long stint as a robotics trainee at Nigeria's foremost robotics and AI research center — Robotics and Artificial Intelligence Nigeria (RAIN). At RAIN, I worked with Dr. Olusola Ayoola on varied projects spanning robot navigation, visual SLAM, and robot control. Before RAIN, in affiliation with the Electrical Engineering Department at my alma mater, I collaborated with Engr. Ihechiluru Okoro on research topics at the intersection of robust control, observer-based compensator design, and feedback control of time-delayed dynamical systems.

I earned my undergraduate degree in Electrical Engineering (with highest honors) from the University of Nigeria, working under the supervision of Dr. Udoka Nwaneto. My bachelor's thesis focused on model-based controller design for speed regulation in electric drives.

profile photo

Contact Info

4448 A.V. Williams Bldg.

8223 Paint Branch Dr

College Park, MD 20740

Email LinkedIn
Research

I study contact-rich poly-articulated dexterous grasping and manipulation under uncertainty, connecting CBF-based safe control, risk-sensitive planning, latent uncertainty estimation, and safe policy learning to build robotic manipulation systems that handle uncertain object pose, friction, contact modes, and target geometry, as well as potentially-destabilizing wrenches that arise from whole-body motion or other task-space or exogenous sources. I validate these ideas in simulation (MuJoCo, Drake, Isaac Sim) and on real robot platforms (see Robots).

Click to enlarge.

Safe, Quality-Preserving Dexterous Grasping

SE(3)-Equivariant, Reach-Aware Grasp Synthesis

VR-assisted teleoperation data collection demo

Safe Teleoperation & Manipulation Policy Rollout

Imitation-Learned Dexterous Manipulation

Publications and more →

News

[preprint] Our new study on robustness preservation in imitation-learned dexterous manipulation under execution speed variation reveals hidden policy fragility that typical success-based evaluations miss.
[preprint] We introduced a CBF-CLF-based control framework for safe and planner-free dexterous grasping with safe-set forward invariance and grasp quality preservation: Grasp Distance Fields.
[paper] Our paper introducing a CVaR-based belief-space MPPI algorithm for safe control has been accepted for presentation at CDC 2026. Links: demos, arXiv.
[preprint] New preprint on FIRMGrasp introducing a novel grasp quality metric (ε(β)) for robust grasp synthesis under uncertain friction. Grasps certified by ε(β) achieve a 70% lift success rate under lateral pull, compared to 25% for grasps certified by the nominal Ferrari-Canny margin but are rejected by ε(β).
[paper] VNB-Grasp, our variational neural belief framework for robust dexterous grasping under visuo-tactile uncertainty, has been accepted to IROS 2026. Links: code, arXiv.
[model] If you work on learning-based dexterous manipulation, have a look at EquiDexFlow, our SE(3)-equivariant flow-matching model for synthesizing contact-aware multi-fingered grasps. To our knowledge, EquiDexFlow is the first learned dexterous grasp generator to produce articulated hand configuration, fingertip contacts, contact normals, and friction-constrained contact forces jointly within one end-to-end SE(3)-equivariant VAE architecture.
Older News
[paper] Our paper on quantile-based distributional RL for safe control was accepted for presentation at CDC 2025.
[epoch] I passed my research proposal exam and advanced to doctoral candidacy.