Genisom AI debuts scalable quadruped at ICRA

A 30 kg payload and five hours of runtime prove mass production is real.

At ICRA 2026 in Vienna, Genisom AI rolled out its M1 industrial quadruped and the L1-series as mature, mass-produced systems. The company frames this as a leap from lab demos to real world deployment, a stance backed by numbers it has been quietly gathering since its December 2023 founding in Beijing. The M1 is rated for a 30 kg continuous walking payload, with a payload-to-weight ratio described as approaching 1:1. It carries an IP67 protection rating and can run up to five hours depending on payload and operating conditions. The platform’s rugged chassis, on board processing, and field-ready design are meant to support more than moments of demonstration and enable sustained operations in industrial settings.

A critical enabler for that performance is Genisom AI’s in-house P85MAX-S joint actuator module, which delivers up to 180 N·m peak torque in an 86 mm diameter form factor weighing about 1 kg. The company says developing actuators in-house gives it direct control over a key hardware-software integration point that shapes locomotion performance, payload capacity, and system reliability. That level of vertical integration marks a deliberate choice to tie mechanical strength to software control loops, motion planning, and perception, a strategy that can shorten development cycles and improve reliability in the field. The broader signal is a push to create platforms that are not only capable on a test rig but also predictable when faced with the messy realities of deployment.

Genisom AI says it has already produced and delivered more than 10,000 units, a milestone the company reports positions it ahead of many early-stage physical AI ventures still chasing pilots. In that sense, the Beijing-based startup frames itself as following a path similar to other practitioners moving toward scale, yet with a sharper emphasis on industrial deployment. The company notes that its M1 and L1-series are designed around manufacturable platforms that can move beyond demonstrations to serve real customer needs. The emphasis on production-ready design and a hands-on hardware-software integration strategy helps explain the company’s confidence in meeting field demands.

From a practitioner’s vantage point, several concrete tensions and expectations emerge. First, the in-house actuator approach can pay dividends in control and reliability but concentrates risk in a single supply chain node for a critical component. If heat, wear, or failures surface in field use, Genisom AI will need robust service and spare-part support to keep a fleet of deployed units healthy. Second, the 1:1 payload-to-weight ratio implies any growth in payload will demand more energy or smarter power management, a constraint to watch as use cases evolve beyond current demonstrations. Third, the IP67 rating signals readiness for dusty, wet, or rugged environments, yet long-term field reliability will reveal maintenance needs that laboratory testing may not capture. Finally, the production milestone raises expectations for a scalable service ecosystem: predictable maintenance, reliable spare parts, and performance consistency across tens of thousands of units will determine whether the platform can sustain a broad industrial footprint.