Agibot G2 vs Figure AI: Inside the Humanoid Robot Factory Livestream That Settled the Lab vs. Real-World Debate

For six days in late June 2026, anyone with an internet connection could watch a fleet of wheeled humanoid robots do real factory work — not in a staged lab, but on a live tablet mass-production line in Nanchang, China. The event, a global livestream by Chinese robotics company Agibot, was widely seen as a direct challenge to U.S.-based Figure AI, which had earlier broadcast its own humanoid demonstrations from a controlled lab environment.

The question the stream aimed to answer: which company is truly ready for commercial deployment? Here is what Agibot showed, how its G2 robot stacks up against Figure AI's Figure 03, and what the broader strategic picture looks like.

What the June 2026 Livestream Actually Showed

Agibot launched its six-day global livestream from June 23 to June 28, 2026, broadcasting from Longcheer Technology's Nanchang factory in China . The feed showed multiple G2 wheeled humanoid robots operating autonomously on a live tablet mass-production line, performing quality-inspection and sorting tasks at multimedia-integrated testing (MMIT) stations alongside human workers .

Key on-screen metrics reported during and after the stream included:

• Up to 310 units per hour (UPH) throughput per robot station
• 18–20 second cycle time per unit
• 99%+ success rate on inspection and sorting tasks
• Line integration in 36 hours
• ~3,000 units processed per shift per robot station

Agibot's central claim was that these robots were operating in a real, active mass-production factory — not a staged lab environment — making the livestream a direct rebuttal to Figure AI's earlier controlled-lab broadcast .

Agibot G2 vs. Figure AI: Head-to-Head Comparison

The two companies represent fundamentally different design philosophies for humanoid robots. Agibot's G2 is a purpose-built industrial wheeled robot optimized for factory throughput, while Figure AI's Figure 03 is a more general-purpose bipedal walking platform designed for eventual home use .

Dimension	Agibot G2 (June 2026)	Figure AI (Earlier 2026)
Venue	Live consumer electronics factory floor (Longcheer, Nanchang)	Lab/headquarters facility (San Jose, CA)
Duration	6 days continuous streaming	8-hour shift (May 13) and a 200-hour earlier run
Task	Tablet inspection, precision fixture insertion, sorting at MMIT stations	Package sorting, sheet-metal parts handling at BMW plant
Robot type	Wheeled humanoid (G2) — omnidirectional base	Bipedal walking humanoid (Figure 03)
Autonomy	Multi-robot fleet, alongside human workers	Single/multi-robot autonomous shifts with Helix-02 VLA model
Commercial deployment	Already deployed in Longcheer's mass-production line since April 2026	Pilot at BMW Spartanburg (30,000 X3 vehicles, 1,250+ runtime hours)
Known throughput	~310 UPH, 18–20s cycle, 99%+ success	~84s total cycle-time target, 37s handling time per part

Key difference: Agibot stressed that its livestream came from a real high-volume manufacturing line already in production, while Figure AI's demonstrations — though technically impressive with bipedal locomotion and the Helix AI model — were conducted in lab or pilot settings . Figure AI's Figure 03 is a bipedal platform with home-use ambitions, whereas the G2 is a wheeled industrial robot optimized for factory throughput .

Technical Specifications of the Agibot G2

The G2 is built for industrial-grade precision and continuous operation .

Specification	Value
Height	180 cm (can lower to ~122.5 cm)
Weight	~185 kg (with batteries)
Degrees of Freedom	26+ total; 7-DoF force-controlled arms
Gait type	Wheeled, omnidirectional (not bipedal walking)
Max speed	1.5 m/s
Payload capacity	30 kg
Battery life	~4 hours per charge (1,652 Wh); dual hot-swappable batteries for 24/7 operation
Sensors	3D LiDAR, RGB-D cameras, force-torque sensors, IMU, 360° perception
AI integration	Multi-modal perception-to-action stack, VLA model
Ingress protection	IP42-rated (dust and splash resistant)
Precision	±0.5mm placement accuracy for fixture insertion

The G2 uses 7-DoF force-controlled arms with integrated force-torque sensing, allowing it to handle delicate electronics components without damage . It is designed for automotive-grade industrial standards and intended for 24/7 continuous factory operation .

Production Metrics and Scale

Agibot's manufacturing ramp has been aggressive:

• 5,000 units rolled off the Shanghai production line by December 2025
• 10,000 units produced by March 2026 — doubling output in three months
• Agibot declared 2026 as "Deployment Year One", shifting from R&D demos to large-scale commercial rollout
• The company introduced an L1–L5 autonomy framework, claiming current systems at Level 3 (conditional automation in constrained environments)
• Agibot stated a target of 10,000 humanoid shipments for 2026, backed by aggressive data-collection expansion to improve real-world performance
• The Longcheer deployment was called the "world's first embodied AI deployment" in a live consumer electronics mass-production line

Broader Strategic Ambitions

"Deployment Year One" (2026): Agibot's leadership explicitly framed 2026 as the year the company moves from lab demonstrations to large-scale commercial deployment of embodied AI across multiple manufacturing verticals . The Longcheer tablet line is the flagship reference case.

Industry vertical expansion: Prior to the tablet factory deployment, Agibot demonstrated G2 in auto parts production (seatbelt lock cylinder assembly, material handling) and precision task workflows . The company aims to penetrate automotive, consumer electronics, and general manufacturing sectors.

L1–L5 autonomy roadmap: Agibot publicly defined a capability ladder with Level 3 as current state (conditional automation in bounded factory zones), targeting higher levels for less constrained environments .

Data at scale thesis: Agibot has emphasized that high-quality real-world operational data — not just hardware — is the core competitive moat, and is aggressively scaling data collection from deployed production units to feed its AI training pipeline .

Production capability as a differentiator: Having already surpassed 10,000 units produced, Agibot claims a massive manufacturing volume advantage over most competitors, which it argues enables faster iteration cycles and lower per-unit costs .

Global PR battle with Figure AI: The June livestream was explicitly timed and framed as a response to Figure AI's lab-based demos, with Agibot asserting that "real factory deployment" versus "corporate lab demonstration" is the meaningful benchmark for the industry .

The Bottom Line

The Agibot G2 livestream marked a shift in the humanoid robotics narrative from "can it work?" to "is it working in production?" While Figure AI continues to advance bipedal locomotion and general-purpose AI, Agibot's bet is that the first-mover advantage in real factory throughput — at 310 units per hour with 99%+ reliability — will define the near-term competitive landscape. Whether that bet pays off depends on how quickly both companies can scale and iterate. For now, the cameras in Nanchang showed one version of the future: humanoid robots earning their keep on a live assembly line.