How robot fights
actually work

Robot fighting is not one sport

The phrase robot fight describes a visual outcome, not a shared sporting contract. A wheeled machine built to disable an opponent, a small biped judged on falls and a full-size humanoid scoring effective strikes may all appear in the same search result while competing under incompatible rules.

Project H therefore starts with the competition file, not the highlight. The operator, hardware class, control mode and scoring document determine what viewers are actually watching. Without those four fields, a knockout clip is entertainment evidence—not a comparable sporting result.

• Identify the competition
• Identify the control contract
• Identify the hardware class
• Find the published resolution rule

Sources S1S2

First ask who makes the decisions

Autonomous and teleoperated bouts can look almost identical after the bell. The important difference sits upstream. URKL’s published rules prohibit joysticks, remote controls and motion controllers during competition, and Chinese reporting describes in-bout decisions as AI-led. REK instead connects a human pilot through VR and motion capture, while UFB uses remote gamepad control.

That distinction changes the meaning of speed, defence and recovery. In a pilot-led format, performance belongs to a human-machine pair. In an autonomous format, perception, planning and control software become part of the competitive athlete. A combined ranking would erase the skill being tested.

Sources S1S3

Hardware decides what is possible

A standardised platform narrows the contest toward software and preparation; an open field makes hardware engineering part of the result. URKL centres its tournament on EngineAI’s full-size T800 and restricts core modification. REK has used Unitree G1 and H1-class machines. UFB has presented a broader mix of Unitree, Booster and related platforms.

Size matters just as much as brand. ROBO-ONE has supported humanoid combat for decades, including an autonomous category, but its robots operate around a four-kilogram class. It is historically essential and technically distinct from a full-size humanoid league. “Autonomous” alone is not enough to make two records comparable.

Sources S1S3S4

A fall, a knockout and a points win are different records

The URKL V1.0.0 rules publish a points ladder for effective actions, a ten-second knockout count, a limit of two resets and an eight-metre competition circle. Tournament bouts are described as best-of-three rounds with five minutes of net time. Those mechanics are what makes a result checkable rather than a matter of opinion.

The Shenzhen opening gala used five exhibition rounds and ended 3–2 on points. That result is documented, but it should not be projected onto the later tournament format. Project H keeps exhibition records, tournament rules and future schedule claims in separate fields precisely because sports data becomes misleading when formats are flattened.

Sources S1S2S5

How a result becomes trustworthy

Video can prove that an event occurred and show a visible incident. It cannot, by itself, supply a judge scorecard, confirm a roster or resolve an unpublished round. The strongest record combines a pre-existing rules contract, an organiser or institutional result and independent reporting that names the same participants and outcome.

When one of those layers is missing, the interface should say so. Matador’s 3–2 gala win is confirmed across Chinese and English reporting, but no source publishes the individual round winners or judge totals. A trustworthy database preserves that absence rather than drawing a plausible-looking score timeline.

• Rules define the possible outcomes
• Named sources resolve the actual outcome
• Footage documents context
• Missing detail remains visibly missing

Sources S2S5S6