Independent Variable has secured nearly 1 billion yuan in Series A+ financing, with participation from Alibaba Cloud, National Science Investment, China Development Bank Capital, Sequoia Capital, and Yingze Capital.

Independent Variable Robotics recently completed a nearly 1 billion yuan Series A+ financing round. This round was led by Alibaba Cloud and Guoke Investment, with follow - on investments from China Development Financial, Sequoia China, and Yingce Capital. Existing shareholder Meituan Strategic Investment made an over - subscribed follow - on investment, and Lenovo Star and Legend Capital continued to increase their investments.

This is Alibaba Cloud's first investment in an embodied intelligence company. Alibaba Cloud highly recognizes Independent Variable's technology and will provide multi - dimensional support. The funds will be used for the continuous training of Independent Variable's fully self - developed general embodied intelligence foundation model and the R & D iteration of hardware products.

Since its establishment at the end of 2023, Independent Variable has established a technical path to achieve general embodied intelligence through an end - to - end unified large model. Recently, it released a fully self - developed wheeled bipedal humanoid robot adapted to multi - modal large model control: Quanta X2. The company's technical route of synchronous development of software and hardware, as well as its forward - looking technical concepts and achievements, have also been recognized by national - level investment platforms, top domestic and foreign investment institutions, and industrial capital.

As the first domestic company to achieve an end - to - end embodied intelligence large model, Independent Variable independently developed the "WALL - A" series of VLA (Vision - Language - Action) operation large models to build a unified cognition and action framework. In a unified representation space, the model processes perception, reasoning, and action simultaneously, directly conducts cross - modal causal reasoning and action decision - making, enabling robots to think and work like humans eventually. Currently, the "WALL - A" model has demonstrated zero - shot generalization ability in some new task types that have not been trained at all.

Meanwhile, the company was the first to achieve an end - to - end embodied thinking chain reasoning framework. Based on multi - modal input, it conducts in - depth reasoning and generates multi - modal output, forming a complete closed - loop of the model's autonomous decision - making, execution, exploration, and reflection. The model can closely combine language understanding, visual perception, and action execution, forming a reasoning process closer to human thinking, successfully breaking through the bottleneck of multi - step long - sequence tasks, significantly improving task completion rates, and greatly expanding the ability boundary of robots to handle complex real - world scenarios.

In the middle of this year, the company was the first to achieve the control of a high - degree - of - freedom dexterous hand for complex operations using an embodied intelligence large model. Previously, Independent Variable released a video of a self - developed large model controlling a high - degree - of - freedom dexterous hand to delicately pick up and distribute elastic and deformable objects such as playing cards.

Currently, to promote the research and application of embodied intelligence large models, Independent Variable has open - sourced its embodied foundation model for developers: "Wall - OSS" and made the relevant training codes public to facilitate global developers to quickly fine - tune and apply it on their own platforms.

Wall - OSS has strong generalization and reasoning abilities. It performs better than other foundation models in long - range operation tasks. As a multi - modal base, the model also has good causal reasoning, spatial understanding, and reflection abilities.

Wall - OSS is an open - source embodied foundation model trained on a large - scale real - world dataset. In terms of model architecture,

it innovatively designed a "Shared Attention + Expert Diversion (FFN)" architecture to transfer the knowledge of VLM to the operation model without loss, achieving a deep coupling of language and action. In terms of training methods, it pioneered a three - stage training paradigm of "discretization first, then continuity, and then combination" to ensure that the cognitive ability of VLM is stably and losslessly transferred and extended to physical actions. In addition, the unified cross - level thinking chain realizes forward arbitrary mapping at the cross - level abstraction level, enabling the model to seamlessly switch between high - level decision - making and low - level execution within a single differentiable framework.

In terms of hardware, in August this year, Independent Variable released a fully self - developed wheeled bipedal humanoid robot "Quanta X2". In less than half a year, the company achieved full - stack self - development of the robot body, high - degree - of - freedom dexterous hand, and exoskeleton remote - control data acquisition equipment.

Quanta X2 is a general robot body native to the model. In its design, it not only considers the needs of model training and complex operation tasks but also achieves a comprehensive balance and optimization in core indicators such as load capacity, operation space, movement speed, and control accuracy.

The five - finger dexterous hand of Quanta X2 adopts a bionic structure design, with 20 degrees of freedom per hand and the ability to sense subtle pressure changes. At the same time, based on the arm - hand integrated exoskeleton technology, Independent Variable pioneered an industry - leading integrated whole - body remote - control solution of "humanoid robotic arm + high - degree - of - freedom dexterous hand". Quanta X2 can not only collect high - quality data to feed back model training but also be deeply integrated with self - developed models, truly entering real - world scenarios for application.

With the improvement of the comprehensive software - hardware integration ability, Independent Variable's robots have reached cooperation with leading service and industrial customers and have been put into use in multiple scenarios. In the future, Independent Variable will also build an open ecosystem with customers around models and hardware to promote the further development of embodied intelligence.

This article is from the WeChat official account "Independent Variable Robotics". Author: Towards General Embodied Intelligence. Republished with permission from 36Kr.