Let's compare
R1-Lite
Galaxea AI
Not yet assessed
- Height
- —
- Payload
- —
- Verified autonomy
- not assessed
- Real deployment
- not assessed
- Status
- —
- Price
- —
R1-Lite
Galaxea AIThe R1-Lite is a humanoid robot with 23 DOF, featuring 6 DOF arms (600 mm reach, 3 kg rated / 5 kg max payload), 3 DOF torso (0–1.7 m vertical range), and 6 DOF chassis (1.5 m/s speed, omnidirectional). It includes an Intel Core i9-12900HK, 32 GB RAM, 1 TB SSD, platform binocular camera, optional wrist depth cameras, and supports R1 Lite Teleop remote operation.
Availability
Specification
- 23 total DOF including 6 DOF arms, 3 DOF torso, 6 DOF chassis
- Arm payload 3 kg (rated) / 5 kg (max) at 600 mm reach
- Torso vertical range 0–1.7 m, with front-back translation
- Chassis speed 1.5 m/s, omnidirectional movement (translation, spin, Ackermann)
- Intel Core i9-12900HK, 32 GB RAM, 1 TB SSD
- High-definition binocular camera on platform, optional wrist monocular depth cameras
- Supports R1 Lite Teleop isomorphic remote operation platform
Price
No public price — contact the supplier for a quote.
Good · Bad · Ugly
Evidence-graded claims from the Galaxea AI deep report
Galaxea AI has raised approximately 3 billion yuan (~$420M+) across multiple rounds by April 2026, reaching a $2.9 billion valuation — making it one of the most heavily funded humanoid robotics startups in China.
Multiple independent financial news outlets including Caixin Global [17] and CnTechPost [6] corroborate the funding rounds and valuation trajectory with consistent figures across Series A, B, and B+; the investor roster including Ant Group, Meituan, and Baidu Ventures is independently named across sources.
from Galaxea AI deep report →
Galaxea R1-series robots are capable of autonomous task execution (manipulation, inspection, logistics) via the G0/G0.5 VLA stack, achieving 89.1% on RoboTwin and 97.0% on LIBERO benchmarks.
Benchmark scores are reported in Galaxea-authored arXiv papers [18][21] and not independently replicated by third-party evaluators; the 97% LIBERO score contrasts with a 48% LIBERO-PRO score, suggesting generalization remains incomplete.
from Galaxea AI deep report →Fast-WAM world model compresses single-step inference latency to 190ms — over 4x faster than traditional approaches.
This latency figure and the 4x comparison baseline appear only in vendor-sourced descriptions [7][11]; no independent benchmark or third-party test confirms the 190ms figure or defines the comparison baseline.
from Galaxea AI deep report →Galaxea plans 10,000-unit mass production of R1-series robots in 2026, with Lens Technology as its hardware supply chain and mass production partner.
The Lens Technology partnership is confirmed by an independent news source [9][17], but the 10,000-unit production target is a vendor forward-looking claim with no independent verification of manufacturing readiness or order backlog to support it.
from Galaxea AI deep report →The R1 Pro humanoid features 20 DOF, 10 kg dual-arm peak load, 500 TOPS onboard compute, and a four-fingered dexterous hand with 17 active DOF and up to 1 kg per-fingertip force.
Hardware specifications are sourced from commerce/product pages [4] and a company video [22], with no independent third-party teardown, test report, or regulator certification confirming these figures.
from Galaxea AI deep report →
G0.5 achieves zero-shot generalization — handling unfamiliar objects, novel scene layouts, and new language instruction combinations without task-specific fine-tuning.
The company's own research papers [18][21] describe a deployment protocol requiring 1 human demonstration plus 40 real samples for stable deployment, directly contradicting the zero-shot claim; the 48% LIBERO-PRO score further indicates significant generalization gaps.
from Galaxea AI deep report →Galaxea R1-series robots are commercially deployed at scale, with 40+ clients including Volkswagen, Huawei Cloud, Haier, Samsung, and ByteDance, selling since late 2024.
An independent news source [9] explicitly characterizes current deployments as pilot-stage and contrasts Galaxea unfavorably with Unitree, which has shipped products to paying customers at scale; the 40+ client figure likely reflects pilot/research agreements rather than commercial-scale shipments.
from Galaxea AI deep report →G0.5 ranks top-1 in China across 7 major global benchmarks, placing it in the global first tier of embodied AI models.
No independent source in the dossier verifies this ranking claim; it appears only in vendor-produced materials [7][11], and the research papers confirm strong but not independently validated top-1 performance.
from Galaxea AI deep report →