EEmbodied AI Hub
IndustryProductFeaturedEditor’s pick

Ecovacs Bypasses Humanoid Hype to Open-Source Modular Home Manipulator

Ecovacs has open-sourced its Bajie physical intelligence platform, including hardware and software, signaling a strategic pivot from end-to-end humanoid VLA models to modular, localized home-space management. The move highlights a growing industry consensus that commercial viability in domestic environments requires pragmatic form factors and community-driven data co-creation.

open-source roboticsmodular platformhome robotEcovacsqbitaiBajiechina

Source: 量子位 · July 27, 2026

Share this article so more people can see it

Ecovacs, the Chinese robotics company best known for its floor-cleaning robots, has taken an unusual step in the crowded embodied AI space: it open-sourced its Bajie (Eight Realms) physical intelligence platform, including the full hardware design, software stack, and AI agents. The move is a deliberate pivot away from the humanoid robot hype that has dominated headlines, and toward a more pragmatic, modular approach to home robotics.

The company announced the launch of the "Bajie Open-Source Innovation Space" in Suzhou, China, where developers, researchers, and students can access the platform for free. The platform is built around the Bajie robot, a wheeled base with a six-degree-of-freedom robotic arm, a silent chassis, and a two-finger gripper with UV cleaning. The robot lacks legs, a face, or any human-like features—a conscious design choice.

"When the concept of robot first emerged, it was meant to do hard labor. Hard labor doesn't have to be human-like," said Ecovacs Chairman Qian Dongqi in an interview. "We focus on what the robot can do in the home, not what it looks like."

The Bajie robot is equipped with dual RK3588 SoCs, enabling it to run a 7B-parameter large language model locally without cloud dependency. Its arm has an 800mm reach, and the body can lift 44cm to access high and low spaces. The platform includes hardware components like the arm, gripper, UV cleaning unit, and cargo tray, as well as software including the underlying system source code, AI agents OpenClaw and Hermes.

Ecovacs has designed four access paths for different user types: natural language commands for non-coders, C++/Python SDKs for developers, SSH/serial tools for researchers, and a full open-source base for enterprises to build commercial applications. The company emphasizes that the platform is not simplified—it retains full robotic capabilities but at a cost accessible to schools, startups, and individuals.

Qian is skeptical of the prevailing VLA (Vision-Language-Action) end-to-end model approach. "Home environments are infinitely varied; a single general model cannot cover all situations," he argued. Instead, Ecovacs advocates a "thousand faces for thousand homes" strategy, where the robot's basic capabilities are first built solidly, then customized per household through community-driven data co-creation.

The open-source strategy is not purely altruistic. Ecovacs acknowledges that no single company can solve the data and adaptation challenges alone. By opening the platform, it hopes to build an ecosystem where developers contribute to improving the robot's environmental understanding and task execution. "This is not charity; we have core interests at stake," Qian said.

The Bajie robot targets what Ecovacs calls "pioneer users" from its existing floor-cleaning customer base. Initial use cases include toy sorting, shoe organizing, laundry assistance, object retrieval, and desk tidying—all spatial management tasks. The company has a long-term roadmap from "tool" to "butler" to "companion," with Bajie covering the tool-to-butler transition.

For founders and operators, the key signal is clear: Ecovacs is betting that the path to commercial home robotics lies not in humanoid form factors but in functional, modular, and open platforms. The company's 14-year history in home service robots gives it unique insights into real-world household needs—insights that are now being shared with the community. The question is whether this open-source approach can generate enough community momentum to compete with well-funded humanoid startups.

Source: 量子位.

Related resources on this hub

Jump to projects, models, or datasets mentioned or closely related.

Discussion

Tell us what you think — comments make stories more useful for builders and founders.

Tell us what you think!

Ecovacs Bypasses Humanoid Hype to Open-Source Modular Home Manipulator

Have an account? Log in to use your display name and avatar.

Email is optional and never shown on the page.

More insights

IndustryProductFeatured雷峰网

Xingchi Power CTO on PULSE: Dual-Track Architecture for More Human-like Home Robots

Xingchi Power's CTO Li Da introduces PULSE, a dual-track architecture that models both physical environment and user state (intention, belief, preference) for home robots. The system prioritizes understanding humans over pure physical intelligence, using end-cloud collaboration and native privacy design to enable personalized, context-aware behavior.

Read
IndustryProduct雷峰网

Tianli Qiming's Education AGI Whitepaper: From Answer-Response to Mind Simulation, Breaking the 'Impossible Triangle'

Tianli Qiming's whitepaper proposes a Cognitive World Model (CWM) to simulate student thinking, moving beyond superficial answer-response AI. The framework, validated across 107 schools and 250,000+ users, aims to solve education's quality-cost-scale trilemma. For founders, this signals a shift from content delivery to cognitive simulation as the next battleground in EdTech.

Read
IndustryProduct雷峰网

Zibian Robot Brings Embodied AI to APEC: The Real Path to Serving People's Livelihood

Zibian Robot, the only embodied AI company invited to APEC's Digital Week, showcased its robot-in-home service and logistics sorting deployments. Founder Wang Qian proposed opening more livelihood scenarios for AI, advocating open-source collaboration. The company's WALL-B world model powers robots that have served over 1,000 households and achieved 1200+ items/hour sorting efficiency.

Read