Electric Sheep Verdie robot uses large world models for autonomous landscaping

Listen to this article

Verdie uses AI to edge and trim lawns and bushes and to blow leaves. | Source: Electric Sheep

Electric Sheep Robotics Inc. today launched Verdie, a new robot using its proprietary artificial intelligence and software. It said it aims to be the first large-scale outdoor maintenance company powered by AI and robotics.

The San Francisco-based company said Verdie can autonomously edge and trim lawns and bushes, as well as blow leaves. Electric Sheep added that its AI agent, ES1, enables Verdie and its RAM lawnmowing robot “to operate in any outdoor setting with zero teaching.”

“The debut of our Verdie robot is the first AI robot for tasks like trimming and edging in the world of landscaping, and it’s exciting to see our ES1 technology power multiple robots that can work alongside a crew without an engineer on-site setting a specific path for them,” said Nag Murty, co-founder and CEO of Electric Sheep, in a release. “We will be rolling out the Verdie to our customer sites throughout 2024 and continuing to build out this fleet of robots as autonomous agents trained on outdoor services.”

“Verdie is inspired by robots such as WALL-E, R2-D2 and BB-8 — moderately complex, non-humanoid agents that perform meaningful work with embodied AI,” said the company in a blog post.

ES1 operates like ChatGPT but for spatial reasoning

Verdie and RAM use AI to understand the lawns around them and efficiently care for them, said Electric Sheep. Based on recent advances in generative AI, ES1 is a learned-world model that enables reasoning and planning for both robots.

To execute tasks like mowing and inventory management, ES1 needs to understand the semantics of the world, create a map that can be used for coverage planning, and highlight the edges of the workable area, the company explained. ES1 can do all these things through dense prediction of a world state with a single model, claimed Electric Sheep.

ES1 is similar to ChatGPT, but it works with spatial AI instead of language, it said. In developing its robots, Electric Sheep said it needed to give them complex, real-time reasoning that can generalize across changes in the environment.

“To tackle this, we utilized two techniques: our foundation world model ES-1 and reinforcement learning (RL) in simulation,” it explained. “ES-1 provides an agent with concepts needed for outdoor work; and has been trained from our robot mowing fleet on thousands of diverse properties.”

“Given this robust representation, we can then perform RL- which teaches the agent to solve a specific new task,” said Electric Sheep. “We found that RL on top of our world model can 1) enable simulation based training of new policies such as string trimming and 2) be generalized to work on a diverse number of sites across the country.”

ES1’s model for reinforcement learning and robot training. Source: Electric Sheep

Verdie, RAM learn to handle tools 

Electric Sheep also needed to enable Verdie to handle tools, so it turned to simulation. 

“For a specific example of this, we consider the task of string trimming, and train a small fully-connected policy on top of the learned embeddings using RL,” said the company. “The policy takes as input time-series data via ES-1’s projected embedding space, then outputs velocity commands for the motors to track at 10 hz.”

“We used NVIDIA’s ISAAC simulation for all training,” it added. “The reward function was set to follow the perimeter of the property with the tool tip of the trimmer. Since our models are already designed to run on Jetson platforms, we can train the entire policy on a single desktop GPU.”

Electric Sheep said it designed both of its robots to start working once they’re on a property and turned on. As a result, Verdie can start working “out of the box,” said the company.

Because the robots don’t need on-site engineers to operate, they can simply be shipped to a campus, homeowners association (HOA), or park and begin tasks alongside the crew. The company credited its full-stack data channel and the large volume of data that the robots are continually trained on. 

Electric Sheep is currently running the ES1 agent on a fleet of 40 RAM robots in hundreds of yards across North America. It said it plans to deploy Verdie with customers in the second quarter of 2024. 

Electric Sheep touts business model as differentiator 

Electric Sheep said it has acquired traditional outdoor service providers to progressively transform operations by deploying its software and robots. Electric Sheep acquired two of these landscaping companies in October 2023, and then acquired two more just a few months later in December. 

The acquisitions are also a way to acquire data that the company can use in a reinforcement learning operational sandbox. In that sandbox, Electric Sheep can build on its foundational model and apply it to its Verdie and RAM robots, according to Murty.

“We are building an RL reinforcement learning factory to train autonomous AI agents to do sustainable outdoor work,” he said.

The acquisitions also allow Electric Sheep to start making money from Day 1, as it operates existing businesses with existing customers. The company, which has funding from Tiger Global and Foundation Capital, has grown its revenue by eight times since implementing this model.

Electric Sheep asserted that its growing pipeline of interested businesses will enable it to grow by 10 times in 2024.

Learn from Agility Robotics, Amazon, Disney, Teradyne and many more.