Hi
Most Edge AI projects fail not during inference, but in the real world—where sensor drift and environmental noise corrupt your data.
This is the final challenge we solve in our last cohort.
For
the last 3 cohorts of our Edge AI on RISC-V workshop, we're focusing on a capstone project you won't find elsewhere: building a self-calibrating sensor hub that uses on-device AI to detect and correct for sensor drift in real-time.
This is the difference
between a theoretical model and a production-ready system.
Here is the advanced technical workflow you will implement:
- Sensor Fusion & Drift Injection: You will fuse data from multiple sensors (e.g., IMU, temperature) on the VSDSquadron PRO board. We will deliberately inject synthetic drift into the data stream to simulate real-world decay.
- On-Device Anomaly
Detection: You will deploy a lightweight Autoencoder model to learn the normal "healthy" sensor baseline directly on the RISC-V core. This model will run continuously, reconstructing input data and calculating the reconstruction error.
- Real-Time Correction Loop: When the error exceeds a threshold (indicating drift), your system will trigger a correction routine. This involves dynamically adjusting sensor readings or recalibrating based on the other, stable
sensors—all without human intervention or a cloud connection.
This is next-level embedded intelligence. You will be working with:
- TensorFlow Lite for Microcontrollers
- CMSIS-NN library for RISC-V optimization
- VSDSquadron PRO emulator (not need for real board)
- C++/Python for deployment and control logic
The skills you gain here in building resilient, self-monitoring systems are what separate a junior developer from a senior architect.
This is your final chance to join. Registration for the final 3 cohorts closes in 3 days. The next offering will be for next quarter.
Do not miss the definitive workshop on
production-grade Edge AI.
Click here to enroll and secure your spot now
Best regards,
The VLSI System Design Team
