.png)
Kameleon
A Breakthrough Towards Autonomy
Kameleon automates the process of selecting sensors for a vehicle in the context of object detection in its surroundings, determines the optimal number of sensors, and evaluates the robustness of sensor configurations. The system is based on machine learning techniques utilizing deep neural networks with attention mechanisms. This enables the fusion of multimodal sensory data to create an accurate environmental map around the vehicle. It represents a step towards vehicle autonomy—from cars to satellites.
Distinctive features
Detection accuracy
Improving the accuracy of object detection in the vehicle’s surroundings.
Data fusion
Enhancing detection accuracy and robustness (e.g., against noise) in the vehicle's surroundings.
Optimization time
Reducing the time required to determine the vehicle’s sensor configuration.
System Advantages
- Optimization of sensor setup for vehicle perception, minimizing sensor costs while maximizing the robustness of sensor configurations against external conditions.
- Assessment of sensor (signal) relevance, optimizing the required number of sensors through:
- Data-driven algorithms (attention modules)
- Correlation with perception quality
- Redundancy management
- Evaluation of data fusion algorithm robustness against noisy and low-quality data through:
- A noise simulator with various distributions
- Quantitative, qualitative, and statistical assessment
The system was developed as part of the project:
"System for automatic optimization of vehicle sensor configuration, reliability assessment, and multimodal sensory data analysis using machine learning techniques (KAMELEON)" (Application No. WND-RPSL.01.02.00-24-0ACE/20-002).
See our other initiatives
John Doe
CEO, XYZ Corp
Jane Smith
CTO, ABC Inc
Mike Johnson
COO, DEF Corp
Sarah Williams
CFO, GHI Inc
Super-resolution reconstruction
The aim of the super-resolution reconstruction (SRR) techniques is to improve the quality and increase the resolution of images (upscale images) while restoring as many details as possible from the source image. Additional goals include: preserving sharp edges, limiting the number of unwanted image distortions after the transformations, producing a visually appealing image.
Φsat-2
- The Φsat-2 mission demonstrates advanced AI capabilities in Earth observation through a 6U satellite equipped with a multispectral instrument, launched in August 2024.
- KP Labs developed a CNN-based algorithm for automatic cloud detection, optimizing data transmission by filtering out cloud-covered images before downlink.
- The project implemented 8-bit quantization to ensure algorithm efficiency within CubeSat hardware limitations while maintaining effective real-time image processing capabilities.
ORCHIDE (Orchestration of Reliable Computing on Heterogeneous Infrastructures at the Edge)
- The ORCHIDE project addresses satellite communication limitations by developing an edge computing ecosystem for autonomous data processing and AI integration in space.
- The system implements a software framework for in-orbit reprogramming and deploys machine learning models directly on satellites, enabling autonomous operations without constant ground contact.
- The solution focuses on real-time decision-making and decentralized computing architecture, supporting extended missions through enhanced system autonomy and resource optimization.
Space-Based Data Centers
- The project addresses the challenge of efficiently handling growing satellite data volumes, as transferring all data to Earth for processing leads to delays and high bandwidth costs.
- KP Labs developed autonomous, in-orbit data processing systems for the Space-Based Data Centers project, designing distributed architectures for large-scale data storage and processing directly in space.
- The project validated machine learning models for space-based systems, explored innovative uses in space exploration, and enabled real-time processing of Earth observation data, reducing latency and bandwidth requirements.