Galbot ET1

Galbot ET1 is a bipedal humanoid robot introduced by the Chinese robotics company Galbot at the 2026 World Robot Conference in Beijing. Galbot presented ET1 on August 19, 2026 as its first bipedal model and as a hardware platform for its AstraBrain-Agent embodied-agent architecture.^[1][2][3]

ET1 is distinct from the company's Galbot G1, which has a humanoid upper body on a wheeled base, and from the heavy-load Galbot S1. At launch, Galbot had not published an ET1 data sheet, price, ordering process, production schedule, or delivery date. Public information therefore establishes an unveiled research and demonstration platform, not a commercially available product.^[1][2][3]

Unveiling

Galbot unveiled ET1 during the World Robot Conference, which opened in Beijing on August 19, 2026. Xinhua's Economic Information Daily reported the model among a group of Galbot robots shown in household, retail, and industrial demonstrations. China Daily described ET1 as Galbot's first bipedal humanoid and photographed it beside the wheeled G1.^[1][3]

The company's official YouTube channel released a 49-second video titled "Meet GALBOT ET1." on August 20. The video is promotional footage and contains no technical specifications or test protocol.^[4] Other footage from the conference showed ET1 performing a street-dance sequence that included floor-supported and inverted movements alongside a human dancer. Reports attributed claims that the robot generated or tracked motions in real time to Galbot and its founder, rather than to an independent test.^[2][5]

Scroll sideways for more →

AstraBrain-Agent

Galbot describes AstraBrain as an embodied AI system intended to coordinate robots with different bodies. Its public demonstrations at the conference included bipedal ET1 units, wheeled G1 units, and heavy-load S1 units. The company says AstraBrain supplies higher-level environmental understanding and task planning while body-specific control systems execute movement.^[1][2]

For ET1, Galbot named the relevant architecture AstraBrain-Agent. According to the company, it combines continuous observation, spoken-instruction interpretation, behavior planning, and motion-sequence generation for real-time interaction. Galbot also says people can teach the robot new movements through interaction without first supplying motion-capture recordings or prepared action videos.^[2][3]

These descriptions are company claims about the architecture. Galbot had not released an AstraBrain-Agent paper, checkpoint, training corpus, source code, evaluation suite, or ET1-specific technical report by August 28, 2026. The public demonstrations do not show how much of a behavior was generated online, how much was learned before an event, or whether unseen tasks can be acquired reliably. They also do not establish continuous real-time learning during ordinary deployment.^[2][3]

Public demonstrations

World Robot Conference

At the launch event, ET1 appeared in a dance performance with a human dancer and demonstrated bipedal whole-body movement. Chinese reporting described eye contact, small head movements, responses to language, and dance movements that included hand-supported floor work and an inverted pose.^[2][5] These observations document a stage demonstration. The associated claims that the routine had no preset script and that ET1 learned or reproduced the dancer's movements in real time came from Galbot and were not accompanied by an outside test protocol.^[2][5]

Some conference reports grouped household demonstrations such as folding clothes, washing, and serving food with the broader Galbot display. They did not always identify which robot body performed each task. Those demonstrations should not be assigned to ET1 merely because ET1 appeared at the same exhibition. Galbot's wheeled G1 and other robots were also present.^[1][2][3]

Tennis demonstration

ET1 units appeared in a tennis exhibition during the opening ceremony of the second World Humanoid Robot Games on August 22, 2026. CGTN published photographs identifying a Galbot ET1 practicing with a technician on August 21 and ET1 units practicing before the event. Its report described a Galbot robot playing singles and doubles with human participants, moving into position and returning shots.^[6]

Galbot characterized the match as autonomous and later claimed more than 100 consecutive rallies. CGTN repeated the company's account that a higher-level system assessed the game while a lower-level controller coordinated walking, balance, and racket movement.^[6][7] No independent technical audit, machine logs, rally definition, miss rate, intervention record, or controller configuration accompanied the public demonstration. The event supports that ET1 robots physically participated in a live tennis exhibition, but it does not independently validate Galbot's autonomy or record claims.^[6][7]

Galbot's earlier LATENT research provides relevant background but should not be conflated with ET1 documentation. The LATENT paper describes learning tennis skills from imperfect human motion fragments and reports deployment on a Unitree G1 robot, not on Galbot ET1.^[8] It does not establish that the conference ET1 used the same policy, training data, or exact control stack.

Relationship to other Galbot robots

Before ET1, Galbot's most visible robot was the wheeled G1, used in the company's retail, pharmacy, and manufacturing demonstrations. G1's wheeled base and lifting torso suit structured indoor work, while ET1 uses legs and was introduced through athletic and interaction demonstrations.^[1][3] Similar names used by other manufacturers, including Unitree G1, refer to unrelated robots.

Galbot also displayed its S1 heavy-load robot at the same conference. Reporting about tasks assigned to G1 or S1, or results obtained with a Unitree G1 research platform, is not an ET1 specification. ET1's height, mass, joint count, actuators, sensors, payload, battery capacity, runtime, walking speed, onboard computer, and manufacturing status remained undisclosed in accountable primary or event reporting reviewed through August 28, 2026.^[1][2][3]

Evidence limits

The launch evidence consists mainly of company material, stage demonstrations, event reporting, and state-media coverage that relies in part on Galbot descriptions. No independent laboratory evaluation, repeatability study, safety assessment, or production inspection was available. Public footage can verify that a physical biped moved, danced, and participated in a tennis exhibition under event conditions. It cannot by itself determine whether behaviors were fully autonomous, learned during interaction, robust to unplanned conditions, or transferable to other tasks.

Galbot's statements about autonomous learning, unscripted interaction, and future deployment should therefore be read as the company's technical and commercial claims. No announced price or availability means ET1 should not be treated as a shipping product, and demonstrations at two controlled events do not establish readiness for unsupervised work around people.^[2][6][7]

References

1. ^Yan Lei. "Galaxy General releases bipedal humanoid robot at WRC." Economic Information Daily, Xinhua, August 19, 2026. jjckb.cn/...c
2. ^Xinhua. "2026 World Robot Conference opens in Beijing; Galbot releases bipedal humanoid robot." August 21, 2026. xinhuanet.com/...c
3. ^Bai Shuhao. "Galbot unveils sporty new humanoid model." China Daily, August 21, 2026. global.chinadaily.com.cn/...87ef90a3106bc57421cc5e
4. ^Galbot. "Meet GALBOT ET1." YouTube, August 20, 2026. youtube.com/watch
5. ^Wang Miao. "Galbot's first bipedal robot Galbot ET1 makes its debut." IT Home, August 20, 2026. ithome.com/...247
6. ^CGTN. "Humanoid robot takes on tennis, testing AI's next frontier." August 23, 2026. news.cgtn.com/...ng-AI-s-next-frontier-1PQ73bW4mSQ
7. ^Beijing Galbot Co., Ltd. "World's First Autonomous Humanoid Tennis Match: GALBOT Robots Complete 100+ Consecutive Rallies." Company press release distributed by PR Newswire, August 23, 2026. prnewswire.com/...00-consecutive-rallies-302857808
8. ^Zhikai Zhang, Haofei Lu, Yunrui Lian, et al. "Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data." arXiv:2603.12686, March 13, 2026. arxiv.org/...2603.12686