Amazon foundation model for robots shows what's possible
The vendor created the new model with Amazon SageMaker, showing what's possible with the platform and that models do not have to be language models to provide business value.
Amazon has rolled out a new foundation model for its entire fleet of warehouse robots.
The vendor on Monday introduced DeepFleet, a generative AI (GenAI) model that can coordinate the movement of robots across Amazon's fulfillment network. This improves the travel time of Amazon's robotic fleet and enables faster delivery of Amazon packages. The new foundation model was built using Amazon's data sets of inventory movement, on Amazon SageMaker. SageMaker is AWS' cloud service for building, training, and deploying machine learning and GenAI models.
Ease and business value
The release of DeepFleet represents GenAI vendors' increasing ease in building foundation models that are not language models, according to Rowan Curran, an analyst at Forrester. It also highlights the expanding value and use cases for the models.
Amazon is not the only vendor creating foundation models for new applications. In March, streaming provider Netflix released a foundation model for recommendations that aims to get information from members' watch history and the provider's content. In February, Microsoft also introduced Muse, a GenAI model for gameplay ideation.
It seems to be much easier, at least as reported thus far, to manage these models when you have one unified kind of training and architecture.
Rowan CurranAnalyst, Forrester Research
"As the maturity of the actual pretraining of foundation model starts to reach a level where companies can form more of these operational types of applications ... we're getting to the point where [foundation models are] becoming much more useful to business," Curran said.
He added that the Amazon example shows that as long as businesses have the massive amount of data and strong training capabilities needed, it's possible to create use cases from foundation models.
"It seems to be much easier, at least as reported thus far, to manage these models when you have one unified kind of training and architecture," Curran said.
The Amazon example also demonstrates what is possible for other use cases. For example, if Amazon can apply the model to help route its robotic fleet, it's possible the company and others can apply it to traffic patterns, cars, planes or boats.
Amazon SageMaker application
"DeepFleet is Amazon's ... implementation of what you can do with SageMaker AI," Nicholson said. "Every AWS sales team that sells SageMaker AI has as their No. 1 case study a million robots."
The challenge is that while Amazon found great success with SageMaker, other AI vendors, such as Nvidia, could argue that they can also provide better success with their tools, Nicholson continued. Moreover, just because Amazon was able to create a large foundation model with SageMaker does not mean that it will work for every enterprise.
"If AWS ... is allowed access to data in a way that no other private company would allow access to their data to AWS, then maybe it's not an appropriate comparison in my business versus Amazon," Nicholson added.
Moreover, Curran said that even large enterprise-level manufacturing or distribution companies might find it hard to build a foundation model that could differentiate because it might not have the scale for data like Amazon.
"Particularly if we see an Amazon or something like that release an open-weight version of something like DeepSeek that could accelerate other work ... in these types of spaces," Curran said.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
