Autonomous Agriculture Engineering

Agriculture Autonomy Provider Cuts Perception Latency from 2s to 200ms

Embedded engineering that enables real-time perception across the autonomous vehicle stack.

200 ms

Streaming latency

Production

Grade Autonomous stack

End-to-End

Simulation & ground truth

About

A global autonomous vehicle technology provider developing self-driving systems for agriculture lacked the embedded software expertise needed to build and validate a production-grade perception stack on NVIDIA platforms. Veltris filled that gap across BSP, sensor fusion, perception, simulation, and ML-ready data pipelines, reducing sensor-to-viewer streaming latency from 2 seconds to 200ms for real-time field perception.

Real-time perception starts with engineering every layer of the autonomous stack.

Challenge

Building production-grade autonomy across a complex embedded stack.

Developing a perception stack on NVIDIA platforms required specialized expertise across BSP development, sensor fusion, 3D simulation, and ML data pipelines. Without a partner capable of engineering across these layers simultaneously, integration became a constraint on the autonomous roadmap.

PAIN POINTS

Niche embedded engineering expertise difficult to source internally
High-latency sensor-to-viewer streaming
Complex sensor fusion across camera, LIDAR, and GNSS
Simulation and ground-truth infrastructure needed for model validation

Solution

Embedded engineering built for real-time autonomous perception.

Veltris built the edge computing stack on NVIDIA NRU-110V, including BSP, Kernel, DTS, Rootfs, fastRTPS+DDS, and ROS, enabling low-latency sensor data control. Perception software was engineered for surround-view cameras with GNSS, optimizing sensor communication, Ethernet drivers, and time synchronization for high-resolution media streams to the GPU.

A 3D simulation environment combined MATLAB Simulink, Unity3D, Blender, and Unreal Engine with custom plugins, fisheye cameras, depth maps, and semantic and panoptic segmentation. An AWS-based ground-truth and ML data pipeline automated labeling, privacy masking, fruit-ripeness classification, validation, and reporting.

Step Flow
Capture → Fuse → Simulate → Validate → Deploy

From sensor data to validated perception, every layer works together.

WHAT CHANGED IN THE WORK

High-latency streaming → Real-time perception
Disconnected engineering layers → Integrated autonomous stack
Manual data preparation → Automated ground-truth pipeline
Limited validation → Simulation-driven development

WHAT WE DELIVERED

NVIDIA Edge Computing Stack
Perception & Sensor Fusion
3D Simulation Environment
Ground Truth & ML Data Pipeline

Ready to accelerate autonomous systems from perception to production?

Results

200ms latency. Production-grade autonomy. Faster validation.

KEY RESULTS DELIVERED

200ms

streaming latency, reduced from 2 seconds, enabling real-time perception for autonomous field operations

Production

grade autonomous software stack on NVIDIA supporting camera, LIDAR, and GNSS inputs

Complete

simulation and ground-truth framework supporting model training, validation, and feature rollout

Why Veltris

Veltris combined embedded systems engineering, perception software, sensor fusion, simulation, and ML data infrastructure to address the full autonomous technology stack. By engineering across hardware abstraction, real-time perception, simulation, and ground-truth pipelines, Veltris helped move the program from integration constraints toward production-grade autonomous operation.

CApaBILITIES APPLIED

Embedded Systems Engineering
NVIDIA Edge Computing
Perception & Sensor Fusion
3D Simulation Engineering
Ground Truth & ML Data Pipelines
AWS Autonomous Data Infrastructure

The team has been instrumental in supporting our embedded systems development, from preparing and integrating custom Yocto images to resolving complex integration challenges. Their expertise in RADAR driver development, proactive bug resolution, and reliable application support has consistently helped us maintain development efficiency and product stability. They understand the depth of what we’re building and they’ve become a critical part of how we do it.

Director Industrial automation and autonomous systems company

See how embedded engineering can advance autonomous systems.