Robotics Inference at the Edge
This five-day program covers the full pipeline from training embodied AI models in simulation to deploying real-time inference on edge hardware. Participants work hands-on with NVIDIA Isaac Sim for synthetic data generation, apply quantization techniques for edge deployment, and build inference pipelines on Jetson Orin. The program culminates in deploying a working VLA model from simulation to hardware.
Curriculum
What You'll Learn
Embodied AI Foundations
- From language models to action models
- Vision-Language-Action (VLA) architecture
- Sim-to-real transfer principles
Model Quantization for Edge
- INT8/INT4 quantization techniques
- Pruning and knowledge distillation
- Inference benchmarking
NVIDIA Isaac Sim
- Scene creation and domain randomization
- Synthetic data generation
- Robot training in simulation
Jetson Deployment
- Model compilation for Jetson Orin
- Real-time inference pipelines
- Sensor fusion and perception
Advanced Topics
- Multi-robot coordination
- Human-robot interaction safety
- Adaptive behavior and continual learning
Audience
Who Should Attend
Robotics engineers and researchers
Computer vision engineers transitioning to robotics
GCC teams building on NVIDIA platforms
R&D leads evaluating embodied AI
Delivery Format
Format
5-day in-person program with lab.
Prerequisites
Python, PyTorch, basic robotics concepts.
What's Included
Jetson Orin access, Isaac Sim license, certificate.
Instructor
Dr. Sayonsom Chanda
Founder, SARAL
Former researcher at NREL and Idaho National Lab. R&D 100 Award recipient. Leads SARAL's research across Energy, AI, Quantum, and Robotics.