Getting Around and Getting Hold: Locomotion, Hands and Runtime
Whether a humanoid can do your job usually comes down to three things: how it moves, what it can hold, and how long it lasts between charges.
These four topics decide whether a robot is useful in a building rather than in a video. Locomotion figures are quoted on flat, clean, level floors. Hands are photographed holding rigid, matte, well-lit objects. Runtime is measured somewhere between idle and full effort, and the datasheet rarely says where.
They belong together because they compete for the same energy. Faster walking, stronger grip and longer runtime are three ways of spending one battery, and a robot optimised for a demonstration is usually optimised for whichever of them the demonstration showed.
How to use these comparisons
The market mixes several kinds of product. The Unitree G1 is a compact platform whose base and EDU editions differ in ways that matter. H1 and H2 are full-size machines with different configurations and handling demands. 1X NEO is presented as an Early Access home robot with ownership and subscription paths and a managed assistance model. Comparing them is only useful once the buyer has stated whether the goal is research, application development, industrial evaluation or household service.
For every topic below, write the requirement without adjectives such as advanced, safe, fast or production-ready. State an observable condition, a measurement, a tolerance and an evidence source, and treat a blank cell in a seller's response as a condition of purchase rather than an optimistic assumption.
- Speed and locomotion
- Balance and terrain
- Hands and end effectors
- Battery and charging
Speed and locomotion
The question this topic has to settle is published top speed, gait stability, acceleration, route completion and converting locomotion claims into operational throughput. That sounds narrow, but it crosses mechanics, software, service and contract boundaries: a published maximum may describe one option, one motion or a laboratory condition, while a buyer needs to know what the delivered robot repeats, under what limits and with what remedy when the answer differs from the quotation.
Unitree publishes dynamic demonstrations and platform-specific speed figures, whereas NEO's domestic value depends more on safe, predictable movement among people than on winning a peak-speed comparison. This is a procurement distinction, not a claim that one platform is better everywhere.
Minimum evidence packet. repeatable route time, acceleration limits, stops, turns, surface transitions, energy per route, interventions and confidence bounds across many runs. Ask the seller to mark each item standard, optional, unavailable or buyer-supplied.
How this goes wrong. a peak-speed number wins the comparison while real cycles are dominated by cautious turns, resets, charging and human clearance
Balance and terrain
The question this topic has to settle is balance recovery, steps, slopes, floor transitions and the limits of spectacular mobility demonstrations. That sounds narrow, but it crosses mechanics, software, service and contract boundaries: a published maximum may describe one option, one motion or a laboratory condition, while a buyer needs to know what the delivered robot repeats, under what limits and with what remedy when the answer differs from the quotation.
Unitree's dynamic platforms invite athletic comparisons, while a home-oriented NEO must prioritize predictable movement and human-safe recovery; every platform still needs tests on the buyer's actual surfaces. This is a procurement distinction, not a claim that one platform is better everywhere.
Minimum evidence packet. surface coefficients, obstacle dimensions, handrail assumptions, fall boundary, recovery method, repeated trials and logs for every intervention. Ask the seller to mark each item standard, optional, unavailable or buyer-supplied.
How this goes wrong. an edited stair or disturbance video is treated as evidence that an unsupported unit can traverse a different staircase under payload and production software
Hands and end effectors
The question this topic has to settle is grippers, dexterous hands, tactile sensing, payload, durability and matching an end effector to real objects. That sounds narrow, but it crosses mechanics, software, service and contract boundaries: a published maximum may describe one option, one motion or a laboratory condition, while a buyer needs to know what the delivered robot repeats, under what limits and with what remedy when the answer differs from the quotation.
G1 can be quoted with materially different hand options, H1/H2 manipulation depends on configuration and reach, and NEO is designed around household interaction where softness, remote assistance and object variability matter. This is a procurement distinction, not a claim that one platform is better everywhere.
Minimum evidence packet. finger layout, force limits, tactile resolution, replaceable wear parts, calibration time, object test set and repeated grasp results including drops. Ask the seller to mark each item standard, optional, unavailable or buyer-supplied.
How this goes wrong. the buyer pays for dexterity but receives fragile complexity that adds calibration and repair without improving the bounded task
Battery and charging
The question this topic has to settle is battery capacity, realistic runtime, charging, swaps, thermal behavior, degradation and transport rules. That sounds narrow, but it crosses mechanics, software, service and contract boundaries: a published maximum may describe one option, one motion or a laboratory condition, while a buyer needs to know what the delivered robot repeats, under what limits and with what remedy when the answer differs from the quotation.
G1 lists a quick-release battery and an approximate runtime, H1/H2 use larger packs for full-size bodies, and NEO's advertised household duration must be reconciled with task mix, standby, connectivity and service expectations. This is a procurement distinction, not a claim that one platform is better everywhere.
Minimum evidence packet. watt-hours, certified battery documents, workload traces, charger power, swap procedure, cycle-life policy, storage limits and end-of-life replacement price. Ask the seller to mark each item standard, optional, unavailable or buyer-supplied.
How this goes wrong. a nominal runtime becomes the staffing plan even though payload, gait, compute and battery aging reduce useful work between charges
This guide uses manufacturer material checked on August 30, 2026: Unitree G1 product page, Unitree H1 product page, Unitree official store, 1X NEO product and ordering page, 1X legal terms. Prices, configurations, delivery queues, software access and specifications change. Product pages document what a manufacturer currently represents; they are not independent performance tests, and they are not a substitute for the final quotation and terms.
Bottom line
Test on your floor, with your objects, for a full shift. Thresholds, cable trays, ramps, spills and the gap under a roller door are the features that decide deployment, and none of them appear on a specification sheet.
Ask for runtime at the duty cycle you intend, charging time from the state the robot actually reaches, and whether anyone has to be present to plug it in.