Case Studies AMP Robotics: Making recycling easy with reliable annotation quality control

Edit This Case Study Record

	AMP Robotics: Making recycling easy with reliable annotation quality control

AMP Robotics: Making recycling easy with reliable annotation quality control

Technology Category	Analytics & Modeling - Computer Vision Software Analytics & Modeling - Machine Learning Functional Applications - Remote Monitoring & Control Systems
Applicable Industries	Recycling & Waste Management
Applicable Functions	Business Operation Quality Assurance
Use Cases	Machine Condition Monitoring Predictive Maintenance Process Control & Optimization
Services	Data Science Services System Integration Training
Challenge	As important as recycling is to the sustainability of our planet, the challenge of doing it efficiently and cost-effectively has evaded the industry. While manual labor has been the mainstay, there are limitations to human perception that AI can overcome. There is immense potential for profitability by integrating this technology, however, it can only happen with high-quality annotated data being fed to the neural networks. Beginning as a small, cross-functional team, many of AMP Robotics’ team members had to share the load of data annotating. However, as the company began growing, the need for additional help became evident. AMP then took the step that many fast-expanding tech companies do and began hiring temp workers through an agency. “But we found that it was expensive, and the quality of the annotations just wasn’t there,” explains Ben Clint, AMP Robotics Data Acquisition Manager. “College students were brought in and trained, but the problem is that part-time, unfocused work is not the best thing to get really high-quality results.” Read More
About Customer	AMP Robotics is a leader in the recycling technology industry, building AI-based robotics systems that sort recyclable material at a fraction of the cost of current technology. While a student at Caltech, founder Matanya Horowitz recognized both the vast untapped potential of computer vision as well as the gap in innovation within the recycling industry. Founding AMP Robotics was his response. The company is headquartered in Colorado, USA, and has a team size of 101-250 employees. AMP Robotics focuses on leveraging AI and robotics to improve the efficiency and cost-effectiveness of recycling processes, addressing the limitations of manual labor in the industry. Read More
Solution	Initially, AMP Robotics had a data pipeline where the CloudFactory team and another offshore team were doing annotations and reviewing each other's work. That quickly turned into the other team doing the annotations while CloudFactory leads quality control (QC) review of the labeling work. Our partnership has grown even further to include reviewing of model outputs and helping to vet new areas for development. “CloudFactory added value beyond the agreed-upon tasks, helping AMP Robotics self-improve at every step,” explains Clint. In essence, the CloudFactory team served as a springboard to improve processes within AMP Robotics itself. As part of this evolutionary relationship, CloudFactory’s added value didn’t end there. According to Clint, “It’s been good always having a veteran on the CloudFactory data team, and new people coming in don’t take long to get up to speed. We use the CloudFactory team for more than just annotation. We’ve begun using them for one-off tasks as well, like the vetting of new projects.” CloudFactory’s model ensures that regardless of turnover, there will be no loss of quality, context, or knowledge. This stability is what helps our clients maintain their growth momentum. Read More Log in to view content
Contents

Technology Category

Analytics & Modeling - Computer Vision Software

Analytics & Modeling - Machine Learning

Functional Applications - Remote Monitoring & Control Systems

Applicable Industries

Recycling & Waste Management

Applicable Functions

Business Operation

Quality Assurance

Use Cases

Machine Condition Monitoring

Predictive Maintenance

Process Control & Optimization

Services

Data Science Services

System Integration

Training

Challenge

As important as recycling is to the sustainability of our planet, the challenge of doing it efficiently and cost-effectively has evaded the industry. While manual labor has been the mainstay, there are limitations to human perception that AI can overcome. There is immense potential for profitability by integrating this technology, however, it can only happen with high-quality annotated data being fed to the neural networks. Beginning as a small, cross-functional team, many of AMP Robotics’ team members had to share the load of data annotating. However, as the company began growing, the need for additional help became evident. AMP then took the step that many fast-expanding tech companies do and began hiring temp workers through an agency. “But we found that it was expensive, and the quality of the annotations just wasn’t there,” explains Ben Clint, AMP Robotics Data Acquisition Manager. “College students were brought in and trained, but the problem is that part-time, unfocused work is not the best thing to get really high-quality results.”

About Customer

AMP Robotics is a leader in the recycling technology industry, building AI-based robotics systems that sort recyclable material at a fraction of the cost of current technology. While a student at Caltech, founder Matanya Horowitz recognized both the vast untapped potential of computer vision as well as the gap in innovation within the recycling industry. Founding AMP Robotics was his response. The company is headquartered in Colorado, USA, and has a team size of 101-250 employees. AMP Robotics focuses on leveraging AI and robotics to improve the efficiency and cost-effectiveness of recycling processes, addressing the limitations of manual labor in the industry.

Solution

Initially, AMP Robotics had a data pipeline where the CloudFactory team and another offshore team were doing annotations and reviewing each other's work. That quickly turned into the other team doing the annotations while CloudFactory leads quality control (QC) review of the labeling work. Our partnership has grown even further to include reviewing of model outputs and helping to vet new areas for development. “CloudFactory added value beyond the agreed-upon tasks, helping AMP Robotics self-improve at every step,” explains Clint. In essence, the CloudFactory team served as a springboard to improve processes within AMP Robotics itself. As part of this evolutionary relationship, CloudFactory’s added value didn’t end there. According to Clint, “It’s been good always having a veteran on the CloudFactory data team, and new people coming in don’t take long to get up to speed. We use the CloudFactory team for more than just annotation. We’ve begun using them for one-off tasks as well, like the vetting of new projects.” CloudFactory’s model ensures that regardless of turnover, there will be no loss of quality, context, or knowledge. This stability is what helps our clients maintain their growth momentum.

Impact #1	AMP Robotics has expanded from 20 to 50 labels since beginning work with CloudFactory.
Impact #2	CloudFactory reviews have resulted in above 95% accuracy in the labels.
Impact #3	CloudFactory’s model ensures that regardless of turnover, there will be no loss of quality, context, or knowledge.

Impact #1

AMP Robotics has expanded from 20 to 50 labels since beginning work with CloudFactory.

Impact #2

CloudFactory reviews have resulted in above 95% accuracy in the labels.

Impact #3

CloudFactory’s model ensures that regardless of turnover, there will be no loss of quality, context, or knowledge.

Benefit #1	95% accuracy achieved in labeling.
Benefit #2	3x increase in monthly project hours.
Benefit #3	2.5x increase in label classifications.

Benefit #1

95% accuracy achieved in labeling.

Benefit #2

3x increase in monthly project hours.

Benefit #3

2.5x increase in label classifications.

Download PDF Version

Overview

AMP Robotics: Making recycling easy with reliable annotation quality control

Operational Impact

Quantitative Benefit