Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)

Unitree A2 Completes 3.9 km Mountain Course in 43 Minutes 41 Seconds

On a 3.9 km mountain course, the Unitree A2 robot from the Beijing Lingyi Technology team posted the best time among quadrupeds: 43 minutes 41 seconds. Participants had to tackle 3,100 uneven steps and nearly 600 meters of elevation gain. The trial exposed the main problems of field robots: actuator overheating, power throttling by the electronics, and loss of stability after a change of payload.

AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
Unitree A2 Completes 3.9 km Mountain Course in 43 Minutes 41 Seconds
Source: TG: @ai_machinelearning_big_data (Machinelearning). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @ai_machinelearning_big_data (Machinelearning) ↗
Editor-in-Chief
Key metric #1The mountain course tested endurance better than a short demo.
Key metric #2The humanoid Unitree G1 overheated and was pulled from the course.
Key metric #3For a product, cooling, autonomy and stability under load matter.

A robot can walk impressively across a flat stage, and that says almost nothing about its readiness for real work. A mountain course is more honest: steps of varying height, narrow sections, a prolonged climb, heated actuators and a battery that cannot be quietly swapped between takes. This is exactly the test China Media Group set up on Mount Tai. The best time among quadrupeds came from the Unitree A2 of the Beijing Lingyi Technology team: 3.9 kilometers in 43 minutes 41 seconds.

For developers, the trophy itself is not what matters. The competition showed which engineering constraints emerge when a robot works for a long time, carries a load and cannot count on a laboratory floor.

What the Course Actually Tested

Around 90 teams from companies, universities and research groups sent their machines onto a historic mountain trail with 3,100 steps and an elevation gain of almost 600 meters. The organizers split the trials into humanoid speed, quadruped speed, load carrying, exoskeleton-assisted travel, and traversal and route finding.

The winner of the quadruped category covered the route at an average speed of roughly 5.4 km/h. For a human, that is an ordinary brisk walking pace, but for an autonomous machine on a long staircase the task is harder: it must continuously estimate body position, choose footholds and compensate for uneven terrain. An error on one step affects the next, and actuator heating gradually reduces the available power margin.

In load carrying, the Lingsi Intelligent team won, scoring 400 points in 1 hour 51 minutes 29 seconds. This result cannot be compared directly with the speed category: it involved a different scenario, a different load and a different scoring system. Still, taken together, the disciplines give a more useful picture than a single run with no useful work.

Where Robots Started to Break Down

The most revealing part of the competition was not the finish but the dropouts. A development version of the humanoid Unitree G1 overheated on the climb, after which its team pulled it from the course. Among other participants, gear motors heated up, control systems limited power and machines drifted off the route. For robots carrying loads, the gait changed after the cargo was repacked at supply points.

This is a familiar systemic problem. A locomotion algorithm may work confidently for the first ten minutes, but after an hour the actuator temperature, battery charge, mass distribution and surface grip all change. That is why a successful demo does not yet prove a product's reliability. What is needed is long-duration testing, temperature and current telemetry, safe speed reduction, detection of gait degradation and a clear stop mode.

The task proved especially hard for humanoids. The organizers even shortened their route because of the terrain. Two legs make it possible to use infrastructure built for humans, but they reduce the stability margin. The quadruped layout distributes load better and keeps three points of support while the fourth leg is repositioned.

What to Take Into Product Development

Teams building robots for inspection, warehouses, rescue work or delivery should check more than top speed. It is more useful to measure continuous operating time, actuator temperature, frequency of operator intervention, energy use per kilometer and the share of the route completed without falls. For a cargo scenario, different center-of-mass positions and load changes in motion should be tested separately.

The Unitree A2 result does not mean quadruped robots are ready to work independently in any mountains. The report does not make clear how autonomous the control was, how many interventions were allowed or how the machines were serviced along the way. Without this data, the competition cannot be considered a rigorous comparative test of platforms.

But a practical signal is already visible: field robotics is limited not only by the quality of perception and path planning. Cooling, power, mechanical reliability and stability under changing load matter just as much. If a robot is to work beyond the stage, the test must last as long as a real shift.

Story

Key Questions Answered

What the Course Actually Tested?

Around 90 teams from companies, universities and research groups sent their machines onto a historic mountain trail with 3,100 steps and an elevation gain of almost 600 meters. The organizers split the trials into humanoid speed, quadruped speed, load carrying, exoskeleton-assisted travel, and traversal and route finding. The winner of the quadruped category covered the route at an average speed of roughly 5.4 km/h.

Where Robots Started to Break Down?

The most revealing part of the competition was not the finish but the dropouts. A development version of the humanoid Unitree G1 overheated on the climb, after which its team pulled it from the course. Among other participants, gear motors heated up, control systems limited power and machines drifted off the route. For robots carrying loads, the gait changed after the cargo was repacked at supply points. This is a familiar systemic problem.

What to Take Into Product Development?

Teams building robots for inspection, warehouses, rescue work or delivery should check more than top speed. It is more useful to measure continuous operating time, actuator temperature, frequency of operator intervention, energy use per kilometer and the share of the route completed without falls. For a cargo scenario, different center-of-mass positions and load changes in motion should be tested separately. The Unitree A2 result does not mean quadruped robots are ready to work independently in any mountains.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…