By Brianna Wessling | December 11, 2024

The month between Thanksgiving and Christmas is the busiest time of year by far for logistics providers, and this year, it’s even shorter than most peak seasons. With Thanksgiving falling just a week later than usual, automation providers such as Ambi Robotics Inc. need to ramp up operations to help customers ensure packages are delivered before the holiday.

For the past several years, Ambi Robotics has worked to take in as much information as it can on how its robots run so it can roll out improvements to prepare for peak season. The Berkeley, Calif.-based company said its robot can flex to meet demand by increasing utilization. This approach has allowed it to shift from a reactive to a proactive approach to this time of year, according to Jeff Mahler, co-founder and chief technology officer of Ambi.

“Peak season is always very stressful for customers,” he told The Robot Report. “There’s a lot of predictions about what’s going to happen, and you never really know what’s going to happen until the day of, so there’s a lot of uncertainty.”

“There’s some uniqueness to this year, as there is to every year, but there’s continued growth of e-commerce,” Mahler added. “I think Cyber Monday was up something like seven or 8% year over year. Online sales were up more than that on Black Friday. So there’s been big spikes in growth in demand for online sales, which drive a lot of volume that we tend to handle.”

Founded in 2018, Ambi Robotics is developing robots that scale e-commerce operations to meet demand. Its latest system, the AmbiSort B-Series, is a modular parcel induction and sorting system using artificial intelligence.

The sort-to-gaylord system can handle up to 1,200 sorts per hour. In addition, it can be used in various use cases, such as reverse logistics, zone skipping, and AI-vision quality control, making it one of the more configurable systems available in today’s market, said Ambi. Last year, the company won an RBR50 Robotics Innovation Award for the system.

Ambi Robotics touts 2024 accomplishments

Some improvements that Ambi Robotics rolled out this year include:

• Rolling out a new AI model architecture for quality control – detecting multiple items on the handoff platform – that increased sort accuracy
• Improved detection methods for swinging items like large bags to reduce motion speed and avoid mis-sorts, further increasing sort accuracy
• Implemented new control software for the robot arm and gantry, helping to increase throughput
• Refined gantry placement locations to improve the number of items in each sack
• Added 2D barcode support, reducing the rate of failed scans

The company said these changes increased throughput to 500 sorts per hour and brought its sort accuracy to 99.6%. For context, Ambi Robotics was performing 410 sorts per hour during peak season in 2023, and 355 sorts per hour in 2022.

Ambi tracks KPIs for each robot

To determine which areas have the most room for improvement, Ambi combs through key performance indicators (KPIs) for each of its robots. The KPIs include uptime, sorting accuracy, and more.

“We try to tie everything we do back to the customer impact and think about what is our biggest opportunity to affect operations now or in the near future,” said Mahler. “We typically are defining that with metrics.”

“Working backward from these metrics, we can come up with a new goal for a KPI target, something where we think there’s an opportunity to do better, and then really dig into the data and look at what’s limiting that KPI,” he explained.

Sometimes, this process involves looking at changes between individual robots, Mahler said. For example, when it comes to throughput, the company’s average sorts per hour was 410 in 2022.

However, that doesn’t mean that every single robot was sorting at that rate. Instead, some sorted at lower rates, and some even faster than 410 per hour. By examining how these robots are performing differently, and what kinds of packages they might be handling, Ambi can identify where it can make improvements.

“We really drill down, actually, all the way into individual sorting events,” Mahler said. “That’s where we typically try to get to. So the robot handled this specific package, and it took way too long. Or maybe there were these 10 packages where it took too long. We try to get a sense of what actually happened here, and then what are our opportunities to fix it.”

Ambi regularly improves the AI behind its robots

Ambi Robotics is continually improving the AI that runs its robots, not just during peak season. In fact, according to Mahler, the company typically rolls out upgraded AI models, equipped with more recent data, about once a month.

Ambi Robotics’ continuous learning with data, powered by AmbiOS

“We have a continuous learning pipeline where we can sample data from the production operations. We can label that data if needed, and then retrain models on that data, and even A/B test them on various robots in the field,” Mahler said. “So maybe one robot gets one model, another robot gets a different model, to see how they’re performing against each other, and then eventually roll that out to the rest of the fleet. So we periodically run this continuous learning operation. It’s roughly once a month.”

In the future, the company is interested in further leveraging generative AI, something that Mahler sees as a big opportunity for the industry as a whole in 2025. For example, the company is looking into using generative AI to help its robots read product labels when it can’t read barcodes. This would allow it to still sort the item even if the barcode is damaged or occluded.

Improving performance without changing hardware

All of these improvements roll out on Ambi’s robots without any hardware changes. This is important for the company, as it wants to get as many improvements to customers as possible using the hardware they already have.

“We can, on occasion, make minor hardware upgrades to the systems, but it’s a huge operational problem,” noted Mahler. “There are costs associated, there are people traveling the site for installation, and so on. So having the constraint of keeping the hardware the same guides our thinking.”

“We have to focus on making the hardware do what it’s capable of and to its best ability. So there’s sort of fewer areas to focus and go really deep on and start to improve performance with,” he continued.

In particular, Ambi has put a lot of time into improving its gripper technology while still using the same hardware.

The company said that looking forward to changes it will target next year, AI has a lot of potential. It’s interested in further improving its accuracy, with an eventual goal to reach 99.9% pick accuracy.

Ambi also interested in improving some of the infrastructure around the robots. This means creating better notification systems for when something goes wrong.

“Operations is really key to actually delivering that value in times like peak season, and really in industrial settings in general. So we’ve learned so much about how to have a coast-to-coast support operation,” Mahler said. “The intersection of operations and technology also allows going from a reactive environment into more of a proactive environment, where there are automated systems that are getting operations ready to deal with problems.”