Deploying Humanoids at Industrial Scale | Brennand Pierce
Bruno Pires, co-founder and chief robotics officer of Bear Robotics, describes the company’s move from autonomous mobile robots to manipulation and humanoid deployments at the Humanoids Summit.
In this talk
- 0:09
From warehouse pilots to production work
Pires opens with a warehouse task: a robot picks up totes, drives them to bays, and returns. He says the work typically takes 40 to 50 people across shifts, and the pilot took about two weeks to set up. The team aims to expand the deployment, but still considers the system early-stage.
- 1:59
What scaling a robot fleet takes
Pires points to Bear’s nine years of operations, more than 16,000 robots manufactured, over 100 salespeople, and six offices as groundwork for humanoid deployments. He says most of the software is shared across AMRs and humanoids, supported by cloud tools for updates, customer configuration, deployment, and debugging. His timeline: about a year to perform a task, three years to run a shift, and five years to deploy at scale.
- 5:32
Factory readiness and fleet coordination
Pires describes the work required to move from a few robots to contract manufacturing, including automated checks for firmware, software, calibration, and driving. He says roughly a third of software engineers focus on calibration and reliability. For multi-robot work in tight spaces, Bear developed robots that coordinate over mesh networks without relying on Wi-Fi or a central planner.
- 10:57
Navigation, simulation, and a recycling pilot
Bear adapted its navigation stack for large warehouses, where changing layouts and pallets made long-term localization difficult. Pires says the team used simulation and generated assets to train grasping policies, then deployed a bottle-packing system in Germany after about six months of work. It achieved a 99% pick rate and 94% place rate, but errors remained too frequent for the customer, so the pilot used hand-written recovery behaviors.
- 17:34
Toward a shared robotics stack
Pires says Bear optimized its GPU pipeline to run Groot at about 15 hertz and found it performed better than the diffusion or action policies they had tried. The company is working to unify software across floor-cleaning robots, AMRs, and humanoids. It hopes to bring a humanoid that loads and unloads alongside AMRs and people to a customer site within a couple of months.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.