Selim, an embedded software engineer, and Alejandro, a lead mechanical engineer at Singapore-based Menlo Research, discuss the Asimov open-source humanoid with the RoboPapers hosts.
Selim says hardware is relatively easy to reverse engineer and may become a commodity, while software is what scales. Menlo aims to offer a reference design people can fork, rather than manufacture robots at mass scale. The team’s earlier robot work also exposed delays and repair barriers when hardware broke.
The team’s goal is to let people without robotics expertise deploy AI agents in physical robots. They describe Asimov as a platform that could host agents such as Jan or OpenClaw, with cameras and microphones providing perception. They expect to run an agentic environment on a robot within one or two months, while noting the need to manage security risks.
The team says low-level access to firmware and hardware helps them address limits in existing robots, including camera placement, power budgets, and motor timing. Their design uses a motion-control board for more than 25 motors and a Linux media board, with timing modeled in locomotion training. They are exploring a repairable three-finger hand, interchangeable end effectors, and accessible service ports for power, USB-C, and Ethernet.
The team is implementing a remote emergency stop and firmware alerts for motor faults, and tests the same firmware in simulation to study motor failures and delayed responses. They estimate the current full robot bill of materials at $15,000, with a target near $10,000 and possible savings from manufacturing changes. They say Singapore offers convenient access to hardware supplies and manufacturing pipelines in China.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Diana Wolf Torres, robotics newsletter writer at Droids and chief evangelist at Lightwheel, discusses physical AI and robotics at NVIDIA GTC 2026 with host Jonathan Stevens on the Thinking Machine podcast.
