01

What breaks when the environment stops being controlled?

Getting a robot to work once is one problem. Getting it to handle changing objects, terrain and people is another. Start with the part of the system that needs to get better.

02

Capabilities to explore

  • Perception, sensor fusion and motion control
  • Manipulation, simulation and embodied learning
  • Component integration and engineering prototypes
  • Field testing, pilot manufacturing and deployment environments
03

Explore the geography

04

Build and test the hardware around a defined failure

A useful China route may combine sensors, component integration, prototype engineering, evaluation and pilot manufacturing. Haidian has reported evaluation and shared-processing platforms; Yizhuang publishes testing and pilot-service priorities. Ask the platform operator whether its equipment and team can reproduce the failure you need to investigate.

Define the hardware, operating conditions and measurement that a physical test must reproduce. Check the equipment, technician support, safety protocol, data access and scene owner before treating a demonstration as validation. Agree what result would justify the next development step.

05

What is blocking the next version?

Separate perception errors from control errors, integration limits and missing field data. Which failure must the next test reproduce?