For a robot learning parkour, practice makes perfect.

Lifelike mechanisms of the sort made famous by viral Boston Dynamics videos are often trained through simulations that mimic a range of real-world environments and situations, according to researchers at MIT’s Computer Science and AI Lab (CSAIL). But available virtual training data is hard to gather in the sheer quantities needed for such a process.

A team of researchers at the lab are proposing a new system that taps generative AI to solve this problem. LucidSim combines a visual generator, physics simulator, and text-generated prompts to scale up the amount of diverse and realistic simulations on hand to train robots.

“Roboticists are trying to replicate the success of text-to-image models or language models like ChatGPT, but ChatGPT is trained on tens of trillions of tokens,” Ge Yang, a CSAIL postdoc fellow and author on the paper, told Tech Brew. “We’re nowhere close to that amount of data, so that’s the problem—robotics has a data problem.”

Keep reading here.—PK

Electrifying the entire US commercial truck fleet is a massive undertaking that could, per one estimate, require hundreds of billions of dollars in infrastructure investments alone.

California-based EV startup Revoy is betting on the idea that fleet operators working to meet decarbonization targets will buy into what it touts as a simpler solution: electric units that can be seamlessly and quickly swapped in and out of semi trucks, essentially converting them to hybrid vehicles and thereby reducing diesel usage.

Revoy now has some insights into how well its solution works; the startup in October announced its first customer, logistics and leasing provider Ryder System. The companies have been working together since last December, with Revoy supplying its solution to Ryder along a route that starts in Memphis and spans portions of Arkansas and Texas.

Tech Brew recently caught up with Revoy’s founder and CEO, Ian Rust, who told us that the deployment is “really showing off our critical application, which is long haul, but also showing how we can get started at smaller scales.”

Keep reading here.—JG

Those AI copilots that tech companies have dispatched to join the office are getting a promotion.

Agents—autonomous systems that can perform tasks beyond the scope of a chatbot—are fast replacing copilots as the buzzword du jour in the race to outfit generative AI for the workplace.

Salesforce has retooled its Einstein Copilot into a new product called Agentforce, which became generally available late last month. Microsoft Copilot isn’t going anywhere, but the tech giant recently added a studio to create agents, as well as a cast of 10 pre-built agents to serve functions like sales, finance, and supply chain. Plenty of others, from Palantir to Asana, have thronged to the build-your-own-agent space as well.

“In contrast to now-outdated copilots and chatbots that rely on human requests and struggle with complex or multi-step tasks, Agentforce offers a new level of sophistication by operating autonomously, retrieving the right data on demand, building action plans for any task, and executing these plans without requiring human intervention,” Salesforce touted in its announcement.

(“Outdated copilots” could be read as a swipe at Microsoft; Salesforce CEO Marc Benioff derided his rival’s Copilot product and its foray into agents in an X missive last month. Microsoft and Salesforce did not immediately respond to requests for comment.)

A new role: Agents essentially aim to shift the AI from the copilot’s chair to the driver’s seat. Rather than just answering questions or spitting out bits of copy at a human’s behest, they can perform multi-step tasks on their own.

Keep reading here.—PK