Independent editorial summaries of third-party projects. These are not ROBODROMO installations or endorsements. Reported results belong to the specific configuration and conditions described.
FEATURED / AGIBOT × LONGCHEER
From robot body to useful work.
A G2 robot, a tablet-testing task and a jointly developed production process. See what the manufacturer and customer actually report.
310tablets/hour · up to, reported by AGIBOTAGIBOT · 15 April 2026
Cases shown: 7
01 / AGIBOTProduction use reported
AGIBOTManufacturer product image · not the installation
AGIBOT G2
Tablet testing, on a working production line.
Longcheer · Nanchang, China
G2 robots load and unload tablet test stations, then sort the tested devices. Longcheer confirms the move into its production lines.
310tablets/hour · up to, reported by AGIBOT
AGIBOT, 15 April 2026 · customer follow-up, 24 April 2026
Read the evidence+
Reported result
AGIBOT reports up to 310 units per hour, over 99.9% task success and more than 140 cumulative operating hours. The release also reports 36-hour line integration after a four-month development programme. These are separate measures, not a single timed trial.
What the project needed
The customer contributed its production process and test equipment. AGIBOT supplied G2 robots and task software; the teams developed the integration together.
What this does not establish
Manufacturer-reported figures, not an independent audit. Public raw logs, exact intervention rates and full project cost are unavailable. A standard G2 order does not include this completed workstation.
For your project
Plan the test fixtures, equipment interfaces and exception handling alongside the robot.
UnitreeManufacturer product image · not the installation
Unitree Go2 + G1
Finding objects across unfamiliar rooms.
Carnegie Mellon University · USA
CMU describes an object-navigation system tested on Go2, G1 and a custom wheeled robot. The software separates reasoning, route planning and motion control.
190real-world experiments · across three robot platforms
CMU thesis presentation · 11 September 2026
Read the evidence+
Reported result
The university reports 190 real-world experiments across the system and navigation that was four to five times more efficient than its research baselines. The abstract does not give a separate success rate or trial count for each robot.
What the project needed
Researchers supplied the navigation and vision-language software. Robot-specific movement control sits beneath shared planning and reasoning.
What this does not establish
Research results, not an installed warehouse service or a standard Unitree feature. The comparison is with research baselines, not human workers. Edition, additional hardware and commercial support require confirmation.
For your project
Choose development access and a technical team around the task you want to build.
UnitreeManufacturer product image · not the installation
Unitree Go2
A Go2 becomes a research platform.
University of Southampton · UK
Southampton documents its own Go2 setup, including ROS 2 control and a successfully integrated Intel RealSense D435i depth camera.
ROS 2documented control and camera integration
University operating guide · undated; checked 22 September 2026
Read the evidence+
Reported result
The guide identifies a working camera integration and both Wi-Fi and Ethernet control routes. It also records that a separate Livox MID-360 integration was unfinished when documented.
What the project needed
The university configured its robot around ROS 2 Foxy and published operating instructions for researchers. This is evidence of practical use and integration work.
What this does not establish
No commercial inspection results, operating-hour total or cost saving is reported. This historical software setup is not a recommendation to adopt the same version today, or proof of compatibility with every Go2 edition.
For your project
Budget for setup, access, compatible sensors and someone who owns the software configuration.
JD describes a Dutch fulfilment operation serving Hunkemöller, using mobile robots, robotic arms, storage and warehouse software together.
12,000m² warehouse · reported by JD
JD Logistics · 10 November 2021; follow-up 8 April 2022
Read the evidence+
Reported result
JD reported a 12,000 m² warehouse and estimated that average daily order production time had fallen by two-thirds. Its 2022 account confirms goods-to-person AGVs supporting business and consumer orders.
What the project needed
JD provided the warehouse operation and integrated automation. Hunkemöller supplied the retail fulfilment requirement. The result concerns the complete facility.
What this does not establish
A historical supplier report, not a 2026 operating audit. The estimated time saving is not an independently verified productivity or headcount saving. The sources do not identify today’s Auto Sorting Wall or Tote Handling Solution as the installed models.
For your project
Start with order profiles, layout and the required workflow before choosing individual machines.
Integrator MS Electronics redesigned SAHM’s glass-decoration handling around two cobots. A CR10A handles the glass; a second cobot performs rotation.
2cobots · handling and rotation split between them
DOBOT customer case · undated; checked 22 September 2026
Read the evidence+
Reported result
DOBOT says the revised cell met the required cycle time and completed site acceptance and commissioning within a few days. It does not publish the cycle time, reject rate or payback.
What the project needed
MS Electronics designed and integrated the cell. The project included factory acceptance testing before installation at SAHM’s production site.
What this does not establish
This is an industrial-arm case, not evidence that the LUMO humanoid can perform the task. The second robot’s exact model is not identified in the account. Numerical productivity gains are not supplied.
For your project
Splitting a task between specialised machines may work better than asking one robot to do everything.
MikoManufacturer product image · not the installation
Miko 3
Miko 3 joins the learning routine.
Primary School Savremena · Serbia
Savremena describes Miko 3 as part of its educational robot programme, with conversation, quizzes and learning activities alongside other classroom tools.
Miko 3named by the school in its teaching approach
School account · undated; checked 22 September 2026
Read the evidence+
Reported result
The school’s own account identifies the robot and how it is intended to support pupils. It does not publish a controlled learning study, attainment results or a verified number of robots in use.
What the project needed
Teachers and the school provide the learning context. The school describes Wi-Fi-connected interaction and links to its teaching materials.
What this does not establish
School-reported adoption, not proof of improved intelligence, wellbeing or exam results. Any local curriculum integration must be checked separately; it is not a standard feature promised with a UK Miko 3 purchase.
For your project
Choose a supervised learning activity first, then check content, connectivity and the school’s data requirements.
Chen and colleagues · Robotics and Autonomous Systems
Researchers tested an XGO-Rider with a cloud language model, local command checks and bounded movement actions. The paper separates the AI service from the robot’s control.
The abstract reports 17.47 ± 2.72 seconds of interaction latency (20 trials). In a matched 100-scenario comparison, task success was 90%, versus 78% for a detector-and-controller baseline.
What the project needed
The research combines cloud speech and language services, visual detection, a local validator and the manufacturer’s movement SDK.
What this does not establish
Results describe a custom research system. The delay is not the motor-control period. They do not establish commercial uptime, offline capability or superiority to trained robot policies. Evidence checked from the publisher’s abstract, not a reproduced experiment.
For your project
Ask where decisions run and what happens when the network slows down. These details affect the experience as much as the robot body.
Skild liefert die Roboterintelligenz; NVIDIA trägt KI-Infrastruktur und Robotik-Tools bei; Foxconn stellt die Fertigungseinstellung bereit. Der Bericht beschreibt zweiarmige Manipulatoren, ohne ihren Hersteller zu nennen.
Was sie entwickelt haben
Eine Sequenz, die eine Busbar und Limit-Block platziert, befestigt 16 Schrauben und reagiert auf Störungen während der Montage.
Was die Beweise sagen
NVIDIA meldet die Zusammenarbeit und zeigt den Assembly-Workflow an. Auch Skild beschreibt den Einsatz im eigenen Account.
Ein ähnliches Projekt planen
Für eine ähnliche Installation vereinbaren Sie im Rahmen der Projektbewertung die anhaltende Zykluszeit, das erwartete menschliche Eingreifen und die vollen Betriebskosten.
Die Lektion für Ihr Projekt
Ein fähiger Roboterkörper benötigt Aufgabensoftware, Werkzeuge und einen Prozess, der gemeinsam getestet werden kann.
Verschieben von Totes in einen bestehenden Workflow
GXO · 27 Juni 2024; später Agility Aktualisierung
Agility Einsatzfilm. Sehen Sie sich die verknüpften Berichte für den Betriebskontext und die gemeldeten Ergebnisse an. Sehen Sie sich den Source Film an ↗
01Erhalten Sie ein Tote
02Lift und Transfer
03Platz auf dem Förderband
Wer brachte, was
GXO lieferte den Lagerworkflow in einer SPANX-Anlage. Agility geliefert Digit und seine Arc-Flottenmanagementplattform im Rahmen eines Robots-as-a-Service-Vertrags.
Was sie entwickelt haben
Transfer von mobilen Robotern zu Förderern, verbunden mit der bestehenden Automatisierung des Standorts durch Arc.
Was die Beweise sagen
Die Handelsvereinbarung folgte einem 2023 Pilot. Agility Später berichteten mehr als 100,000 bewegte Toten bei GXO Die Einrichtung Flowery Branch.
Ein ähnliches Projekt planen
Bewerten Sie für Ihr Lager die Kosten pro akzeptiertem Tote, Interventionsrate und Leistung im Vergleich zum vorhandenen Workflow. Diese maßnahmen werden nicht durch die kumulative tote-zählung geliefert.
Die Lektion für Ihr Projekt
Beginnen Sie mit einer Übergabe zwischen bestehenden Prozessen und messen Sie, wie zuverlässig es funktioniert.
BMW Gruppe · Retrospektive März; Juni 2026 Folgemaßnahmen
01Einen Teil abrufen
02Stellung in der Leuchte
03Handover zum Schweißen
Wer brachte, was
BMW lieferte den Produktionsprozess und die Werksteams. Figure geliefert Figure 02 und arbeitete mit BMW über den Antrag.
Was sie entwickelt haben
Abholen und Positionieren von Blechbauteilen für das Schweißen im Werk Spartanburg mit Änderungen der Sicherheitsvorkehrungen und Anlagenvernetzung.
Was die Beweise sagen
BMW Berichte über 90,000 Komponenten, die über rund 1,250 Betriebsstunden während seiner 2025 Pilot. Diese Zahlen beschreiben, dass Pilot, nicht jeder Figure Roboter.
Ein ähnliches Projekt planen
Dies war ein dediziertes Projekt mit BMW und Figure Ingenieurteams. Ihre Aufgabe benötigt eine eigene Ausrüstung, Integration, Unterstützung und kommerzielle Bewertung.
Was als nächstes passierte
Im Juni 2026, BMW Die Erkenntnisse aus Figure 02 bildete die Grundlage für eine neue Figure 03 Logistikprojekt. Der Nachfolger hat unterschiedliche Hardware und Fähigkeiten; Das Lernen wurde über Robotergenerationen hinweg fortgesetzt.
Ein fokussiertes erstes Projekt kann Site- und Task-Wissen für das nächste aufbauen. Das ist ein Grund, früh zu lernen, anstatt eine Garantie für zukünftige Einsparungen.
Start with one useful task. We can help shape the brief and identify a supplier or technical partner. Agree the equipment, support and success measures before committing to a trial.
