1X EVE

What EVE is

EVE is a wheeled humanoid robot built by 1X, the Norwegian company long known as Halodi Robotics. It is roughly human-sized and balances on a wheeled base rather than walking on legs. In its March 2023 announcement the company called EVE its "first commercially available android" and described it as a wheeled android^[2].

Where it came from

Wikipedia records that 1X, then named Halodi Robotics, released EVE as its first humanoid robot in 2018, designed for logistics, security and medical environments^[9]. New Atlas first came across a pre-production prototype of the wheeled humanoid in 2019, when Halodi set up a second base of operations in California^[8].

How it is built

A July 2022 EEJournal article described EVE as a self-balancing, human-sized robot with 23 degrees of freedom, force and impedance control on all joints, two-arm manipulation and an 8 kg payload on each arm^[6]. The same article noted that EVE was wheeled rather than legged, that it could squat down to pick things up off the floor, and that it supported a VR avatar mode that let a remote human operator take control^[6].

Those two design choices go together. Balance is a hard control problem for a walking machine, and a wheeled base sidesteps part of it while keeping the upper body — arms, joints, sensors — where the manipulation work happens. Holding the floor-level workspace is then a matter of squatting rather than of stepping.

In April 2024, 1X said that every component of EVE and NEO was designed and produced in-house at its facility in Moss, Norway, and that all its androids feature gearless motors and a soft exterior^[5].

The security contract

The commercial path began with a strategic partnership announced in April 2021, in which Halodi Robotics secured an investment from ADT Commercial to develop a robot complementing its physical security solutions^[1][11]. Halodi then signed an agreement dated 22 March 2022 to provide 140 humanoid robots to ADT Commercial for security on commercial premises, and the company said the rollout was the industry's single biggest contract for humanoid robots^[1][11].

Learning to work

The public record of EVE's later life is mostly a record of learning methods. In a February 2024 demonstration reported by New Atlas and TechInformed, 1X said it had trained 30 EVE robots using imitation learning from video and teleoperation, and that every behaviour in the video was controlled by a single vision-based neural network emitting actions at 10 Hz^[7][10]. The company said the video contained no teleoperation, no computer graphics, no cuts, no video speedups and no scripted trajectory playback, and showed EVE robots opening doors, picking and placing objects and plugging themselves into charging stations^[7][10].

In a 31 May 2024 AI update, 1X said it had built a voice-controlled natural language interface on EVE to chain short-horizon capabilities across multiple small models, and noted a general difficulty: in multi-task models under 100M parameters, adding data to fix one task can degrade others^[4]. The company also stated that the video accompanying that update was not completely autonomous, because humans dictated when the robots should switch tasks^[4].

In a September 2024 post, 1X said it had gathered thousands of hours of data from EVE humanoids doing diverse mobile manipulation tasks in homes and offices and interacting with people, and had combined video and action data to train a world model^[3]. As one test, 1X placed EVE in front of a mirror to see whether the world model would generate mirrored actions, but did not see successful recognition or self-understanding^[3].

Funding and place in the line

1X announced on 23 March 2023 that it had raised US$23.5 million in Series A2 funding led by the OpenAI Startup Fund with Tiger Global, Sandwater, Alliance Ventures and Skagerak Capital, and said it would use the funds in part to scale manufacturing of EVE in Norway and North America^[2].

EVE is therefore the machine on which 1X worked out both a commercial story — a wheeled android sold into security work — and a research story, in which data from a fleet of robots is used to train networks and world models. The company's later statements about in-house manufacturing cover EVE alongside its successor NEO^[5], and the published demonstrations document both what the robot could do and where the learning methods still fell short^[3][4].