Product Release Highlights in Robovision for Q2 2025

Introducing Intelligent Data Capture (IDC): Your Solution for Smarter Data

To help you overcome these challenges and ensure the integrity of your AI models, we're thrilled to introduce Intelligent Data Capture (IDC), a crucial new feature in our Robovision AI platform. This enhancement allows for more targeted and efficient data capture directly from Agents, ultimately facilitating more robust, accurate, and reliable AI models. This feature makes your Agent not just a passive logger but an active data curator, ensuring you quickly gather the right variety of images—edge cases and all—without drowning in irrelevant or duplicate frames.

With IDC, you'll gain unprecedented control over your data collection process:

• Precise Image Storage Criteria: Define specific conditions for how and when images are stored, ensuring you capture exactly what you need.
• Strategic Random Sampling: Continuously capture diverse data, even for challenging scenarios like false negatives. This broadens your dataset for retraining, leading to more comprehensive model performance and uncovering hidden areas for improvement.
• Confidence-Based Filtering: Isolate and store predictions within a defined "grey zone" of confidence levels. This is invaluable for fine-tuning models and optimizing inference thresholds, helping you strike the right balance between minimizing false positives and avoiding missed events.
• Prioritized Class-Based Sampling: Focus your data capture on critical or specific defect classes, ensuring essential data is always fed back for model retraining. This also lays the groundwork for better support of unknown classes in the future.

Our Intelligent Data Capture comes fully integrated with our Tier One  algorithms, and soon also with your custom algorithms, setting the stage for more advanced and future-proof workflows on Agents.   

While data captured via IDC currently resides on the Agent and requires manual transfer to the hub, we are actively developing automated upload solutions for subsequent releases. This initiative is integral to our vision for a more comprehensive Agent management toolset and the eventual realization of truly self-optimizing AI models.

The Cost of Compromised Data

Poor data isn't just inefficient; it can actively undermine your AI investment and/or deployment:

• Inaccurate Predictions and Decisions: At best, a model trained on poor data will make unreliable predictions. At worst, it can lead to deeply flawed and harmful decisions in real-world applications, whether it's misdiagnosing a medical condition, skipping defective parts on a conveyor , or causing a self-driving car to malfunction.
• Wasted Resources: Cleaning and preparing data is already a significant investment of time and resources. If you're working with fundamentally bad data, you're pouring effort into a leaky bucket. This translates to wasted computational power, developer hours, and financial outlay for a model that simply won't perform as expected.
• Lack of Generalizability: A model trained on poor data may perform adequately on the specific, flawed dataset it learned from, but it will struggle once deployed on your production line  when confronted with real-world data. It won't be able to generalize  its learning, rendering it effectively useless outside of its narrow, compromised training environment.

During Vision AI deployment, compromised data erodes customer trust, making the system appear unreliable or biased, and quickly undermines user confidence and C-level support. For system integrators, this reputational damage can trigger customer churn and devalue crucial AI initiatives. 

Ultimately, treating your data with the respect it deserves—ensuring its accuracy, completeness, relevance, and representativeness—isn't just a best practice; it's non-negotiable for building effective and ethical AI. The challenge: capturing enough defect-laden images when modern production lines distribute flaws so sparsely.

Storage Monitoring: Keeping an Eye on Your Edge Devices

For uninterrupted AI operations, understanding and proactively managing storage on your edge devices is absolutely paramount. Our new Storage Monitoring feature provides comprehensive, real-time insights into storage space usage, whether your deployments are on physical edge servers or within cloud-based hubs.

You'll gain immediate visibility into critical storage categories:

• Data Storage: Monitor space consumed by captured images and crucial derived data, such as embeddings for SAM.
• Robovision Database Usage: Track the footprint of the platform's metadata database, essential for system performance.
• Algorithm Space: See the disk space occupied by your deployed algorithms and their Docker containers.

The user experience for storage monitoring has been thoughtfully designed to reflect the underlying infrastructure. On Agents, a unified graph clearly displays overall disk space and the individual contributions of all data types. For cloud-based hubs with separate virtual disks, three separate pie charts (one per each aforementioned storage category) allows for precise, individual disk monitoring.

This powerful feature delivers early warnings and critical alerts as disk space thresholds are approached, empowering you to take proactive measures. You can swiftly decide to delete data, optimize storage configurations, or plan disk upgrades, ensuring continuous operation. While predictive capabilities for future disk fullness are not included in this initial release, this robust monitoring functionality provides the crucial real-time insights you need. We understand the unique challenges of managing remote edge devices, and this enhancement marks a foundational step toward providing even more comprehensive remote management tools for your entire agent ecosystem.

A New Data Center: Your Centralized AI Data Hub

Finally, v5.10 comes equipped with our dedicated Data Center, a critical new feature designed to give you unparalleled control over your AI workflows and data, basically acting as a catalog/library. It addresses the growing need for efficient data organization and management, unifying previously disparate functionalities into one powerful environment giving a far superior way to keep track of interesting images containing ‘outliers’. Can be easily tagged so that they can be picked up by your labelers in the Label Center.

The Data Center provides:

• Streamlined Data Organization: Effortlessly organize all your data in one primary environment. We've consolidated labeling tools, tags, defect books, annotations, predictive labeling, user management, reviews, dataset creation, exports, and charts into a single, intuitive view, eliminating previous noise and bringing clarity to your data assets.
• Enhanced Import Insights: Gain detailed visibility into your data imports. Easily track who imported what, when, and include crucial custom metadata like temperature or location (using the SDK).
• Future-Proof Data Ingestion: The Data Center will become the future destination for images dropped by headless agents via Intelligent Data Capture and recordings from cameras. This integrates seamlessly with inference monitoring, allowing you to specify rules for capturing critical samples when certain events occur (e.g., "killer defects" or unhygienic conditions), ensuring rapid response and model iteration.
• Accelerated Dataset Creation: Get a comprehensive overview of all your datasets, including creation history and usage. Quickly filter and tag interesting samples across various imports, allowing you to focus your labeling efforts and accelerate model improvement through efficient transfer learning.

Incoming - Software Development Kit ‘SDK’

We’ve started renewing our Software Development Kit, with the full release planned for Q3. In the meantime, we’re excited to offer a preview—a selection of new and compelling tools to whet your appetite for what’s to come. Think of it as an appetizer before the main course arrives later this year!  These tools are what we call  Robovision-certified Algorithm Extensions (plugins). 

And where previously, these powerful extensions (like the Wafer Map (see image below) as used in chip making fabs or Model Parameter Optimization) were an indivisible part of existing Robovision AI Solution Flows. 

With 5.10, they are now separate, tested components that can be installed on multiple algorithms, including your own custom-built models. For instance, you can now easily combine our Wafer Map feature with your internally developed custom classification model. 

This distinction between the "Shape  your own solution" using our SDK and our quality-assured, certified extensions provides your Data Scientists with unparalleled flexibility empowering them to close the below “Customer-Specific Gap” in no time.

Optimized Help Center

We've entirely redesigned our Help Center to enhance your experience on the platform. While keeping our familiar top bar, the new intuitive landing page makes it even easier to navigate and understand crucial actions within the AI lifecycle, explained by means of user-flows. Plus, you can easily discover all new features at a glance in the "What's New" section at the bottom, complete with direct links to comprehensive how-to guides. 

Users can access it from the platform by clicking the help icon in the lower-right corner (F1 keyboard shortcut).

 Stay Ahead with the Latest Robovision Release

At Robovision, we're committed to continuous innovation, rapidly integrating the latest advancements in AI to empower your operations. By consistently deploying the newest release of the Robovision platform, you not only gain immediate access to these powerful new features like Intelligent Data Capture  and enhanced Storage Monitoring and our Data Center, but you also ensure your systems remain at the forefront of the rapidly evolving AI landscape. Staying current means your models are always benefiting from our latest performance optimizations, security enhancements, and expanded capabilities, helping you maintain a competitive edge and unlock even greater value from your Vision AI solutions.

Do you need help switching to v5.10? Get in touch by replying to the email you received.