top of page

Control One: Pioneering Vision-to-Action AI for Field-Proven, Slow-Moving Equipment

In the world of industrial automation, Control One is breaking new ground by transforming slow-moving, field-proven material handling equipment into intelligent, autonomous Physical AI agents.

In the world of industrial automation, Control One is breaking new ground by transforming slow-moving, field-proven material handling equipment into intelligent, autonomous physical AI agents. The company has developed a vision-to-action AI model that enables equipment such as forklifts and pallet trucks to operate with minimal human intervention, while still maintaining human oversight where necessary — boosting worker productivity. This advanced system, powered by NVIDIA AI and robotics technologies, represents a significant leap forward in warehouse automation.

A member of the NVIDIA Inception program for cutting-edge startups, Control One innovates centrally on a three-brain system, where AI operates in harmony with human input to deliver the most efficient and safe decision-making processes. This hybrid model allows for scalable autonomy in dynamic environments, such as warehouses, while retaining human control over critical decisions.


The Three-Brain System: A New Paradigm for Autonomous Operations

At the heart of Control One’s innovation lies its three-brain system, where responsibility is divided across three layers:

  1. The Human Brain: Operators define high-level commands and provide strategic direction, such as deciding which goods to move or which tasks to prioritise.

  2. The One AI: This centralised AI brain translates human commands into specific tasks, coordinating the flow of operations and managing fleet how each of equipment is utilised. It enables smooth task assignment and monitors performance.

  3. Retrofitted Machines: Equipped with intelligence, these machines autonomously handle perception, navigation, and obstacle avoidance. They take high-level instructions from the One AI and execute tasks on the ground, adjusting to dynamic changes without needing constant supervision.

This architecture ensures that tasks are delegated to the appropriate "brain" for optimal performance, with humans guiding overall objectives, the AI managing operations, and the machines performing on-site actions.


Transforming Vision into Action: How Control One’s AI Works

Control One’s vision-to-action AI model enables traditional equipment to operate autonomously by converting visual data into real-time actionable tasks. Here’s how the system operates:

  1. Perception: The AI agents are equipped with stereo depth cameras and other sensors to collect real-time visual data. Using computer vision algorithms, the system identifies and classifies objects, detects obstacles, and understands the spatial layout of its environment.

  2. Mapping and Localization: Using Visual Simultaneous Localization and Mapping (VSLAM) algorithms, the system creates a real-time 3D map of the warehouse, enabling machines to localize themselves and adapt to changes in the environment. This is essential for efficient navigation in dynamic settings.

  3. Task Intelligence: Based on the mapped environment and received instructions from the One AI, the machines autonomously execute tasks such as transporting goods, avoiding obstacles, and recalculating paths as needed. This distributed intelligence helps ensure the machines can operate without constant guidance from the central AI.

This vision-to-action pipeline allows Control One’s AI-powered systems to not only perceive and analyse the environment but also autonomously navigate and make decisions, leading to faster and more efficient operations.


Powering Control One’s Autonomous Systems with NVIDIA Technologies

To bring this high level of automation to life, Control One relies on NVIDIA’s Isaac accelerated libraries and AI models running on  the NVIDIA Jetson AGX Orin system-on-module. These platforms provide the necessary computational power, real-time processing, and AI optimization to enable seamless operation in complex warehouse environments.


NVIDIA Jetson AGX Orin: The Core of Real-Time AI Processing

The Jetson AGX Orin platform delivers the computational power required for Control One’s AI agents. With its advanced GPU architecture, Jetson AGX Orin handles the heavy-duty tasks of processing sensor data, running AI models, and performing inference in real time. This allows the machines to operate autonomously, even in complex, dynamic environments where quick decision-making is essential.


NVIDIA Isaac ROS: Optimising Robotic Operations

NVIDIA Isaac ROS, a collection of accelerated computing packages and AI models, provides a robotics software framework that supports key functionalities such as SLAM, obstacle detection, and path planning. With Isaac ROS, Control One’s AI agents can create and update maps of their surroundings, identify potential hazards, and navigate efficiently through warehouse spaces. This helps ensure that the machines can operate smoothly, even as the layout or conditions of the environment change.


NVIDIA TensorRT: Accelerating Deep Learning Inference

NVIDIA TensorRT, a deep learning inference optimization library, accelerates the performance of Control One’s AI models. By optimising object detection, perception, and navigation tasks, TensorRT enables AI agents to operate effectively without delays, make real-time decisions and adapt to their surroundings as needed.

YOLOv8 Performance

YOLOv8 has demonstrated remarkable improvements in terms of accuracy and speed compared with its predecessors. With advanced real-time object detection capabilities, YOLOv8 has become a popular choice in various applications, including robotics and augmented reality.

Model detection block diagram

Implementation


Technology used:

  • YOLOv8, NVIDIA TensorRT, NVIDIA GPU

  • Software: Python, NVIDIA Isaac ROS

Methodology

1. Preparing the custom data:

  • Collect Data: Capture warehouse conditions using a camera.

  • Prepare the Dataset: We used a 1280-pixel resolution for detecting pallets.

  • Annotate the Dataset: We labelled images with Makesense.ai to identify pallets and other objects.

2. Model Selection:

  • YOLOv8l was chosen due to its balance of speed and accuracy, which made it suitable for real-time applications.

3. Environment Setup:

4. Configuring and Model Training:

  • After preparing the dataset, we trained the model using YOLOv8l

  • We updated .yaml files for the number of classes and dataset path

  • The model was trained on 1280 *720 pallet images.

Block diagram for YOLOv8 training on custom dataset


5. Optimization with TensorRT:

Controlone logo_Main Black bg orange.png

Conversion of best .pt model to NVIDIA TensorRT

  • Configuration of the NVIDIA TensorRT engine for optimal performance, considering:

    • Precision (FP16 vs. FP32)

    • Batch size

    • Layer fusion

    • Kernel selection

6. Detection with NVIDIA TensorRT:

  1. OpenCV was used for capturing images, as well as for preprocessing and post-processing tasks.

  2. We ran inference using the NVIDIA TensorRT

Running inference using NVIDIA TensorRT


Results

The results for NVIDIA Isaac ROS object detection with YOLOv8:


Addressing Critical Challenges in the Warehouse Industry

Control One’s AI-powered system helps address some of the most pressing challenges in logistics and warehousing, including labour shortages, safety risks, and operational inefficiencies:

  1. Labour Shortages: Control One's autonomous agents help address the skilled-labor shortage by handling repetitive and labor-intensive tasks, increasing operational efficiency.

  2. Enhanced Safety: Safety is a major concern in environments where heavy equipment is constantly in motion. Control One’s AI systems, equipped with vision-based obstacle detection and real-time navigation, significantly reduce the risk of accidents. Machines can detect potential hazards and adjust their movements accordingly, helping ensure a safer environment for human workers.

  3. Efficiency and Scalability: The modular nature of Control One’s AI solution means that it can be easily scaled across different warehouse operations. By integrating more AI-powered machines and refining their operational strategies, businesses can significantly improve throughput and reduce operational downtime.

Conclusion: Shaping the Future of Warehouse Automation

Control One’s vision-to-action AI and three-brain system are revolutionising warehouse automation. By integrating advanced NVIDIA AI and robotics technologies, the company has created a powerful, scalable solution to help tackle the most critical challenges in logistics. The combination of human oversight, central AI coordination, and autonomous machine execution means that operations are not only efficient but also adaptable to changing environments.

As the demand for smart warehouses continues to rise, Control One is positioning itself at the forefront of this transformation. By enhancing the intelligence of field-proven equipment, Control One is driving the future of AI-driven automation, creating smarter, safer, and more productive industrial environments.

For more details on Control One’s innovative solutions, visit Control One or explore the NVIDIA Inception program.

Do you have questions about

Autonomous AI Agents for your operations

or
Integration of One AI ?

Let’s Meet

OFFICE

Operations Office


7-P, Ittina Nagar, Jigani, Bengaluru, Karnataka 560105

/control-one-ai

COMPANY

ONE Transform Program

In the Media

Customer Journey

INDUSTRIES

Industrial Warehouses

Manufacturers

AMR SOLUTIONS

One AI Pallet Trucks

One AI Forklifts

One AI Reach Trucks

TECHNOLOGY

Control One AI

One AI

C-One Commander

C-One Command Station

© 2024 | Control One Logistics Private Limited  | Privacy  | Terms of Use

bottom of page