Making multi-robot reliability
measurable, comparable, and citable — for the first time.
R-Power designs its own off-road test carriers, vehicle-mounted measurement hardware, and evaluation software — covering the full chain from capture to evaluation to metrics. We provide the yardstick for multi-agent systems, not another vehicle or another algorithm.
Four product lines across the full evaluation chain
Carrier, measurement, judgment, and simulation — four pieces that form one complete evaluation infrastructure.
R-ORV off-road test carrier
304×215 mm footprint · 60 km/h straight-line top speed (35 km/h track average) · ≤¥30K per unit · expendable. Built for adversarial testing, fault injection, and multi-vehicle scenarios.
R-Box measurement unit
Fault injection · unified time base · tamper-evident · 17+ dimensions of stable capture. Measurement is evidence.
R-Eval evaluation platform
Temporal alignment · event segmentation · RRI / CRI metrics · reproducible. Delivered as SaaS or on-premise.
R-SIM simulation platform (planned)
A simulation environment closed-looped with R-Box / R-Eval field data: real fault data feeds back to refine the model, narrowing the sim-to-real gap. Beta in Q4 2027.
Ten thousand simulation runs teach less than one crash in the wild
Real faults don't live in simulation
Wheels stuck in sand, thermal shutdowns, communication drops, multi-vehicle conflicts — a single desert run costs over ¥450K, and simulation can't reproduce it.
Multi-agent coordination has no yardstick
Single-robot perception has benchmarks (nuScenes, Waymo, LynxScenes). Hand-off between two vehicles, fault coverage, degraded communication — still blank.
Research chassis can't take the abuse
Existing research platforms: ≤10 km/h, flat indoor floors, single-robot, ¥75–150K per unit — lab assets too precious to crash.
A simulation + field dual loop
Our development pipeline is built on the NVIDIA robotics ecosystem: real fault data feeds back to refine the simulation model, and every simulation iteration cuts field-test cost.
Batch simulation of powertrain, terrain interaction, and multi-vehicle scenarios
Model quantization and operator optimization for the edge
On-robot edge deployment: real-time inference and capture
Field calibration; measured data flows back to simulation
Hong Kong entity, Shenzhen–Hong Kong collaboration
R-Power Technologies Limited
Hong Kong entity, Registration No. 79936367. Handles international business and financing, physically separated from the non-fundraising mainland entity.
University partnerships
MOUs signed with Northeastern University and Hunan University of Science and Technology — joint research and talent channels running.
Intellectual property
Chinese invention patent application accepted (No. 202611531457.4, pending); service inventions of core inventors assigned to the company.
The real vehicle runs; the concept leads
92 seconds: R-ORV field footage × a concept visualization of adversarial competition.
Footage includes real R-ORV field testing and concept (CG) segments of competition scenarios, original audio.