Brochure

About the course

This 15-hour course fills the skill gap in mid-career automotive engineers and research scientists by teaching how foundation models work and where they apply in the automotive domain — from battery degradation forecasting to ADAS trajectory planning.

The course ends with a capstone project where each engineer writes a deployment proposal for their own product area, leaving with working code, peer feedback, and a concrete next step.

What you learn

Foundations and generative models

Bayesian thinking, neural networks and Transformers as function approximators, and the generative model families (VAEs, GANs, diffusion) that underpin modern automotive validation.

Diffusion, forecasting, and diagnostics

Diffusion mechanics and trajectory generation for ADAS, synthetic scenario and defect generation, time-series foundation models, and anomaly detection for battery, vibration, and CAN-bus data.

VLAs, world models, and deployment

Fine-tuning with LoRA, imitation learning and diffusion policy for robotics, VLA architectures for perception-to-action control, world models for autonomous driving, and integration with classical control and safety certification.

15 lecture topics

  • Probability and Bayesian thinking. Distributions, uncertainty estimation, and Bayesian reasoning applied to equipment sensor noise and failure probabilities.
  • Neural networks and Transformers. Function approximators — from classical PID and physics lookup tables to learned dynamical surrogates.
  • Generative models overview. VAEs, GANs, diffusion, and why physical engineering validation requires synthetic scenarios and stress-test data.
  • Diffusion model mechanics. Forward noising, reverse denoising, and score-based matching for continuous multi-dimensional engineering data.
  • Diffusion for trajectory generation. Diffusion-Planner and score-based trajectory generation for robotic arms, AMRs, UAVs, and warehouse automation.
  • Diffusion beyond navigation. Synthetic sensor data imputation, stress-testing for electrical grids, and manufacturing defect synthesis.
  • Time-series foundation models. TimesFM, Chronos, Lag-Llama, and MOIRAI — architectures and zero-shot forecasting for energy, telecom, and IoT.
  • Anomaly detection and predictive maintenance. Vibration analysis for turbines and rotating machinery, battery thermal runaway precursors, and telemetry monitoring.
  • Fine-tuning for your domain. LoRA adapters, parameter-efficient fine-tuning on multivariate sensor streams, and plant energy management.
  • Imitation learning fundamentals. Behavioral cloning, DAgger, and diffusion policy for precision assembly robotics, bin picking, and automated handling.
  • VLA architectures. RT-2, OpenVLA, pi-zero, and GR00T N1 for vision-language-action perception-to-action control.
  • VLAs across industries. Visual quality inspection in semiconductor fab, warehouse logistics, and multimodal technician co-pilots.
  • World models for engineering systems. Generative physical simulation, digital twins, and learned environment dynamics replacing hand-crafted simulators.
  • The foundation model stack. Integrating diffusion planners, TSFMs, VLAs, and world models with classical control, SCADA, and safety certification.
  • Capstone workshop. Each engineer develops a foundation-model deployment proposal for their own product area and industry domain.

Capstone project

Deployment proposal

Each engineer arrives with a real product-area problem that could benefit from a foundation-model approach.

The workshop. Across the final sessions, participants scope the problem, sketch an architecture, prototype in Python or MATLAB, and defend the design under structured peer critique.

Deliverable. A written deployment proposal, working prototype code, peer feedback, and a concrete next step to bring back to the team.

Foundational course taught to your team at your company’s location.

Send a short note with your team’s product areas, headcount, and preferred dates. We reply within two business days with a scoping call.

abhishek.rimc@gmail.com

Who benefits & how

  • Engineering managers. Understand exactly where foundation models will succeed vs. fail. Cut through the hype to improve resource allocation, reduce risk, and guide product strategy.

  • Research scientists. Gain a common language to communicate across engineering and management, bridging the gap between theoretical AI and concrete business value.

  • Engineers. Move past academic theory to hands-on implementation. Accelerate your career by delivering working code and deployment proposals that directly impact your product area.

Prerequisites

  • Linear algebra background
  • Exposure to probability and optimization
  • In-class coding in Python or MATLAB

Cost

USD 2000 per enrolled student, with a minimum engagement of USD 30,000.

Schedule

Five days, with 3-hour morning lectures (30-min break) delivered on blackboard/whiteboard at the host site. One-hour afternoon office hour every day for questions and application discussions.

Selected recent work

  • Probabilistic contraction analysis of iterated random operators. IEEE Trans. Automatic Control, 2024.
  • Dynamic watermarking for finite Markov decision processes. IEEE Open J. Control Systems, 2025.
  • Robustness to modeling errors in risk-sensitive MDPs. IEEE Open J. Control Systems, 2025.
  • Model-free change-point detection for mixing processes. IEEE Open J. Control Systems, 2024.

Honors

  • Lumley Research Award — OSU College of Engineering, 2019
  • Kenneth Lee Herrick Memorial Award for best PhD student — UIUC Aerospace, 2014
  • Three OSU patents on control and learning systems (two commercialized)

Course principles

  • No hype demos. Every model taught explained with its failure modes. If we can’t state where it breaks, we clearly state it.
  • No black-box library calls. Diffusion, TSFMs, and VLAs are taught end-to-end — noise schedule to sampler. Everyone can derive it on a whiteboard.
  • Every session ends with a stop/go rule. A written decision framework per technique: does it belong in your product area, at what phase?
  • Capstones are production-shaped. Real hardware constraints, real safety constraints, and a real integration path.

Foundation
Models for
Engineering Systems

A 15-hour applied course for mid-career engineers, managers, and research scientists.

The foundation-model stack for engineering systems

Instructor: Prof. Abhishek Gupta

Associate Professor of Electrical and Computer Engineering at The Ohio State University and Founder & CEO of Ensemble Control Inc. (robotics for extreme environments). Author of 70+ publications in leading journals and conferences on reinforcement learning, world models, and stochastic control for autonomous vehicles, energy systems, and robotics. Ph.D. Aerospace (2014, UIUC); B.Tech. IIT Bombay (2009). Three OSU patent applications, two commercialized. Lumley Research Award (2019) for foundational contributions to stochastic control.

Contact Information