MIT researchers have developed an artificial intelligence system that could enable warehouse robots to deftly handle odd-shaped packages and navigate crowded spaces without endangering human workers.
The breakthrough comes as retailers and logistics companies face mounting pressure to automate their operations amid surging eCommerce demand. While robots excel at repetitive tasks like moving pallets, the new PRoC3S technology could finally crack the long-standing challenge of robots safely performing more complex warehouse jobs that traditionally require human dexterity and spatial awareness.
“In theory, PRoC3S could reduce a robot’s error rate by vetting its initial LLM-based assumptions against more specific and accurate understandings of the warehouse environment,” Erik Nieves, CEO and co-founder at Plus One Robotics, told PYMNTS. “Think about it like this: A warehouse robot operating solely on LLM guidance has been described how to complete a task. The PRoC3S concept goes one step further by placing a digital robot in a simulated environment of that task. It’s essentially the difference between classroom instruction and a really good field trip.”
As PYMNTS previously reported, robotics and AI technologies are transforming traditional distribution yards through specialized autonomous vehicles equipped with robotic arms that can handle complex tasks like connecting brake lines and positioning trailers. These robotic systems, which can operate alongside human workers, are helping to modernize a critical supply chain bottleneck where millions of trailers and containers have historically relied on manual, inefficient processes.
AI-driven warehouse robots are advancing logistics by improving efficiency and addressing labor shortages. Agility Robotics’ Digit uses AI to pick and sort in fulfillment centers. Amazon’s Sparrow applies AI for object recognition and sorting, increasing the speed and accuracy of warehouse operations and automating repetitive processes.
MIT’s new PRoC3S system tries to make robots smarter and safer by combining AI language models with computer vision. Before taking action, the robot tests its plans in a virtual environment to ensure they will work in the real world. If a plan isn’t feasible, it tries new approaches until finding one that works.
The system proved successful in lab tests, completing tasks like drawing shapes and sorting blocks with 80% accuracy. This approach outperformed existing methods and could eventually lead to home robots reliably handling complex requests like “make breakfast” by verifying each step virtually.
“LLMs and classical robotics systems like task and motion planners can’t execute these kinds of tasks on their own, but together, their synergy makes open-ended problem-solving possible,” Nishanth Kumar, co-lead author of a paper about PRoC3S, wrote in a blog post. “We’re creating a simulation on-the-fly of what’s around the robot and trying out many possible action plans. Vision models help us create a very realistic digital world that enables the robot to reason about feasible actions for each step of a long-horizon plan.”
The advances in combining AI language models with robotics could help overcome these implementation challenges. Jenny Shern, general manager at NexCOBOT, told PYMNTS that conventional warehouse robotics have been held back by the need to create detailed, step-by-step operating procedures for even basic tasks, making implementation a time-consuming and costly process. This rigid approach has limited robots’ ability to adapt and has increased the resources needed to deploy automation in fulfillment centers.
“MIT’s PRoC3S system aims to leverage advanced vision models and large language models (LLMs) to enable robots to reason about their environment and make decisions at each step of a complex task,” she said. “If this technology is successfully put to use in warehouse robotic systems, it will reduce the need for extensive pre-programming, minimize human intervention and costly errors, and significantly improve operational efficiency, especially in terms of time-cost optimization.”
Current warehouse robots are often constrained by rigid, predefined instructions, which limit their ability to adapt to dynamic environments, Shern said. For example, if instructed to place boxes of varying sizes onto racks, these robots may stop functioning when the first layer is full, unable to decide the next steps.
“It seems that with PRoC3S technology, robots can autonomously assess the environment, identify feasible actions — such as placing items on subsequent layers — and carry out tasks more flexibly,” she said. “This ability to adapt is particularly valuable for unstructured or unpredictable warehouse tasks where numerous variations exist.”
We’re always on the lookout for opportunities to partner with innovators and disruptors.
Learn More