Hillbot

A US embodied-AI startup training robot skills with a mix of simulated and real-robot data.

Hillbot is a US embodied-AI startup whose slogan is “building general-purpose robots one skill at a time.” Its approach combines real-world data with large volumes of synthetic data generated in a simulator to train manipulation skills that generalize; its website shows demos across multiple robot bodies, including an arm opening a cabinet door, a quadruped, and a dual-arm system. Its website lists the open-source simulation platform ManiSkill and the simulation framework SAPIEN as its own products; both projects came out of Hao Su's lab at UC San Diego, and Su is reportedly a co-founder of Hillbot. Its founding date and funding are not disclosed on its website. For newcomers, it's a representative example of the “simulation data camp” among robotics startups.

ExampleLarge numbers of cabinet-opening and grasping demonstrations are first generated in parallel on GPUs inside ManiSkill, then used together with a small amount of real-robot data to train a policy.

Related

ManiSkill · SAPIEN (SimulAted Part-based Interactive ENvironment) · Synthetic Data · Sim-to-Real Transfer · Real-Robot-Data Camp vs. Sim-Data Camp · GPU-Accelerated Parallel Simulation

Sources

Hillbot 官网 (Chinese)
ManiSkill (haosulab) GitHub

As of

2026-09