Krishna Kumar | Civil and Environmental Engineering

Associate Professor

Research Interests

Physical AI & embodied autonomy, Physics-grounded perception & material identification (3D Gaussian Splatting, weak-form mechanics), Adaptive learned simulators & operator learning, Differentiable world models & online control, Robotic construction & soft-material manipulation, and Scientific AI infrastructure for engineering & natural hazards.

Contact

k.kumar@berkeley.edu

Office

643 Soda Hall

Office Hours

Google Scholar

Krishna Kumar is an Associate Professor of Civil and Environmental Engineering at UC Berkeley. He joined Berkeley in January 2027 through the College of Engineering’s Engineering + AI initiative. Prior to joining UC Berkeley, he was J. Neils Thompson Centennial Teaching Fellow and Associate Professor at the University of Texas at Austin and core faculty at the Oden Institute for Computational Engineering and Sciences.

Kumar develops computational methods that help robots predict and act in unfamiliar physical environments. His research combines computational mechanics, machine learning, and robotics, with a focus on deformable materials and complex physical interactions.

His group studies how to infer material behavior from limited observations and use that information to make fast predictions for planning and control. The group develops weak-form methods for identifying constitutive behavior from observed motion, graph-based learned simulators for particle dynamics, and operator-learning methods for efficient simulation. These efforts support a central goal: building world models that adapt their predictions as robots encounter new materials and interactions. Applications include construction automation, soft-material manipulation, navigation on deformable terrain, lunar excavation, and natural hazards.

Kumar directs Chishiki.ai, a $7 million NSF-funded national ecosystem for AI in engineering, and has contributed to NSF’s DesignSafe-CI cyberinfrastructure since 2019. He is a co-founder of Tektos AI, which focuses on construction automation and data centers.

His honors include the Grainger Foundation and National Academy of Engineering Frontiers of Engineering program (2026), the NSF CAREER Award (2024), UT Austin’s Dean’s Award for Outstanding Teaching (2023), the C.S. Desai Biennial Award (2012). He received his Ph.D. from the University of Cambridge in 2015 and subsequently served as a Research Associate and Bye-Fellow at King’s College, Cambridge.

Physical AI, adaptive world models, computational mechanics, learned simulation, in-context learning, differentiable control, and embodied autonomy.

Physics-grounded perception and material identification. Combining 3D Gaussian Splatting, physical constraints, and weakform to infer material behavior from observations. Weak-form mechanics provides efficient constitutive identification and a means to evaluate the physical consistency of learned or generated motion.

Adaptive learned simulators. Developing GNS, operator-learning methods, and reduced-order models that incorporate conservation laws and material-dependent dynamics. A central direction integrates material-conditioned learned simulation and in-context adaptation, allowing world models to generalize to unfamiliar materials from limited interaction data.

Differentiable world models and control. Combining differentiable physics, reduced-order simulation, and function-encoder representations to enable fast prediction and planning in deformable environments. The goal is to adapt both physical models and control strategies online without expensive retraining or repeated high-fidelity simulation.

Embodied autonomy in complex physical environments. Developing physically viable world models for robotic construction, soft-material manipulation, deformable-terrain navigation, and multi-robot coordination. Applications extend to autonomous parafoil flight and lunar construction, where physical properties and environmental conditions must be inferred during operation.

AI for engineering and natural hazards. Building computational methods and open research infrastructure through Chishiki.ai and DesignSafe-CI to advance scientific discovery, simulation, and autonomous engineering systems.

News

No mentions in News

Spotlights

No mentions in Spotlights

Student Updates

No mentions in Student Updates