On Monday 31 August 2026, Singapore's Land Transport Authority put a Unitree humanoid named Olly on the Downtown Line concourse at Little India MRT. The window runs through 11 September 2026, weekdays only, 10:00 to 12:00. CNA's Justin Ong filed a hands-on from the station that evening and updated it at 20:00 Singapore time. Malay Mail / Newswav carried an English amplification on Tuesday 1 September. Straits Times and Mothership had already printed the schedule from LTA's 20 August Facebook post. That is a timed, chaperoned guide trial inside a live station. It is not a rush-hour worker, and it is not a free-roaming platform attendant.
CNA prints the body number Monday's earlier schedule notes skipped: about 1.27 m tall, high-visibility vest, male voice after a wake phrase. CNA also names the maker: Unitree Robotics. Those are opening-day body and vendor claims. They are not a joint continuous-duty table. They are not a payload at reach. They are not a thermal limit for a two-hour off-peak shift under station HVAC and PA noise.
Walking guide. Not a planted counter.
Read the placement carefully. LTA put Olly on the Downtown Line section of Little India and told Straits Times the station has a simpler layout and sits near LTA headquarters, where the machine charges daily. Officers follow it at all times. CNA's reporter watched it turn toward Exit F, ask permission, then walk slowly to the fare gates closest to that exit. When someone blocked the path, it said please stand clear. That is bipedal locomotion in a public concourse with a human safety envelope. Do not confuse it with Monday's other Asian metro story.
Osaka Metro's Hinata trial, which we already covered, is a Gemini-fronted information counter inside Metro Opus shops next to ticket gates. Hinata is installed. Olly walks. Keep those theses apart. One measures dialogue at a fixed desk. The other measures speech, navigation, and gait inside a fare-paid zone with real crowd pressure, even if the clock is only two off-peak hours.
LTA's own bar is modest. CNA quotes the authority: trial during off-peak to see live-station performance and public reaction; no immediate plan to deploy. Straits Times quotes LTA on mapping and familiarisation before and during the window. That is an acceptance pilot. It is not a staffing replacement claim.
Day-one failure modes are the useful data.
CNA's Monday walkthrough is more useful than the announcement copy. Simple service questions worked: first and last train times. Directions were mixed. Tekka Centre took several prompts, then produced verbal directions without a physical lead. On a second try, Olly misheard Tekka Centre as Teacher's Centre and sent the reporter the wrong way. KK Women's and Children's Hospital went better: Exit F, offer to lead, slow walk to the fare gate, then left / escalator / F entrance instructions after the gate. A US visitor who used to work in robotics told CNA the balance looked steady and the walk had no wobble, then asked for a directional microphone because background noise chewed the speech front end.
Those are the failure modes a station engineer actually cares about. Automatic speech recognition under PA and crowd noise. Wrong destination from a near-homophone. Latency of several seconds before a reply. Gait that is safe and slow rather than Games-sprint fast. A stop command when a person steps into the path. None of that is a joint torque curve. All of it beats a press photo of a robot waving at a ribbon cutting.
Unitree catalog numbers are not Monday's datasheet.
Monday's materials do not print a Unitree catalog code for Olly. They do not say G1. They do not say H1. CNA's 1.27 m height sits near Unitree's published G1 standing height of 1,320 mm on the official G1 page, and nowhere near H1's full-size frame. That height rhyme is a filter, not a SKU proof. FIRGELLI will not upgrade a height match into a confirmed model name.
FIRGELLI already hosts a Unitree G1 spec card. On that card and on Unitree's own table, G1 stands about 1.32 m, weighs about 35 kg, claims about 90 N·m at the knee on the base machine, about 2 kg arm payload, a dual-encoder joint stack, local air cooling, and roughly two hours on a 9,000 mAh pack. EDU configs can raise knee torque and arm load on paper. Those are catalog rows for G1. They are not printed on Monday's LTA or CNA copy as Olly's numbers. Do not copy 90 N·m or 2 kg onto the Little India unit without a named SKU.
What stays unpublished for the trial machine is exactly what a buyer would ask first. Degrees of freedom on the deployed build. Peak versus continuous torque at hip, knee, and ankle under stop-and-turn loads. Wrist force if any. Finger DoF. Pack watt-hours and measured runtime after mapping plus dialogue. A continuous-duty temperature limit for a humid underground concourse. A certified stop distance next to a fare gate. LTA's two-hour weekday window and daily charge at headquarters already hint that endurance is a constraint. Hint is not a datasheet.
What this trial can actually prove.
A supervised, two-hour, off-peak guide trial can answer real questions. Do commuters stop. Does the wake phrase work over station noise. Does Exit F guidance beat a wall map for a visitor. Does the machine keep balance when a child walks alongside. Does anyone escalate to the following officer when the model invents Teacher's Centre. Those are useful acceptance metrics. They are not proof that a Unitree humanoid can replace a station assistant at peak, on a platform edge, or during a service disruption.
Keep the adjacent stories separate. Osaka's Hinata is a planted multilingual counter. Galbot's Hong Kong store open is a retail location claim with an unpublished shopkeeper SKU. J-HRTI's Narashino data factory is teleop logging on G1-class bodies. Monday's Singapore story is speech plus slow walking inside one Downtown Line pocket, with LTA saying there is no immediate deployment plan.
If you are comparing this trial to Kinisi's crankshaft bearing pilot or Osaka's counter demo, hold it to the same bar. A named duty. A published force or payload. A measured failure mode. Monday opened a chaperoned walking guide at Little India. It did not publish the joints.