How Humanoid Robots Slashed Marathon Times from 2h40 to 50 min – A Mathematical Analysis

Evolution Curve of Robot Performance

In 2025 the champion humanoid robot "TianGong Ultra" finished a half‑marathon in 2 h 40 min 42 s, while in 2026 the champion "Glory Lightning" completed the same course in 50 min 26 s, surpassing the human world record. The article investigates whether this improvement follows a linear, exponential, or bounded pattern and identifies the underlying engineering advances.

Building a Curve from Two Data Points

With only the 2025 and 2026 results, a simple exponential decay model with a lower bound is proposed. Let t be the finishing time (minutes) for year y . The model assumes: t(y) = L + (t0 - L) * exp(-k*(y - y0)) where L is the physical lower bound, t0 the 2025 time, and k a decay constant.

t0 = 160.7 min (2025 champion)

t1 = 50.4 min (2026 champion)

L = theoretical minimum time (derived from power‑density limits)

Using the human half‑marathon record (57 min 20 s) as a reference for sustainable speed, the physical lower bound speed is estimated, and the corresponding time is calculated.

Solving the equations yields predicted times:

2027: 30.9 min (≈11.38 m/s)

2028: 28.6 min (≈12.31 m/s)

The model quickly converges toward the lower bound, but its limitations are evident.

Limitations of the Model

1. Sample size of only two points – exponential decay is highly sensitive to the endpoints; any race anomaly (robot fall, weather, track differences) would skew the estimate.

2. Diminishing returns – while the 2025‑2026 jump was dramatic, further reductions (e.g., from 50 min to 30 min) will encounter increasingly hard engineering bottlenecks.

The curve reflects the typical early‑stage “explosive gain” phase of a technology S‑curve, where progress slows as physical limits are approached.

Cooling Advances

In 2025 engineers sprayed water on joints to manage heat; in 2026 the champion employed liquid‑cooling originally used in smartphones. Liquid cooling can increase the heat‑transfer coefficient by 5–10× compared with air cooling, enabling joints to maintain thermal balance during continuous high‑intensity running.

Key Technological Breakthroughs

① From emergency cooling to proactive thermal design – liquid‑cooling modules are now integrated, allowing on‑board power without shutdown.

② Whole‑system weight reduction – extensive use of carbon‑fiber and alloy materials cut robot mass by ~15 %, raising acceleration by roughly 17.6 % for the same drive force.

③ Battery and endurance leap – higher‑density cells and improved battery‑management systems extended effective range from ~5 km to >10 km, supporting the 21 km race.

④ Mature autonomous navigation – autonomous teams now handle localization, mapping, path planning, and dynamic obstacle avoidance without human control, comprising about 40 % of entries.

⑤ Scale‑driven component iteration – real‑world race requirements have driven joint lifespan improvements of 2–3× and increased annual shipments of Chinese manufacturers (e.g., Yushu, AGIBOT) to over 1,000 units each.

Is Robot Evolution Faster Than Human?

Human marathon records have improved only 3‑5 % over the past 30 years, whereas robot half‑marathon times improved ~68 % in a single year. Robots are on the steep early part of an S‑curve, while humans are near physiological limits. Engineering constraints can be rapidly overcome with investment, unlike biological limits.

In 2025 China invested roughly ¥73.5 billion (≈$10.8 billion) in humanoid robots and embodied AI, providing the financial foundation for this rapid iteration.

Nevertheless, experts caution that speed in a half‑marathon does not equate to general‑purpose capability; tasks requiring fine manipulation, perception, and adaptability remain far beyond current performance.

Just as Formula 1 spurred advances in automotive braking, cooling, and tires, humanoid robot races are pushing joint, battery, and thermal technologies toward broader industrial adoption.