Independent Variable Robotics has recently concluded an A+ round of financing, raising nearly 1 billion yuan. This funding round was spearheaded by Alibaba Cloud and Guoke Investment, with contributions also coming from Guokai Financial, Sequoia China, and INCE Capital. Notably, existing investor Meituan Strategic Investments substantially increased its share, while Legend Star and Meridian Capital maintained their follow-on investments.

This investment represents Alibaba Cloud's inaugural foray into an embodied AI company. Alibaba Cloud holds Independent Variable Robotics' technology in high esteem and has pledged to offer comprehensive support. The raised funds will be earmarked for the ongoing training of Independent Variable's proprietary general-purpose embodied AI foundational models and the R&D evolution of its hardware offerings.

Since its inception at the close of 2023, Independent Variable has charted a definitive technical course to realize general-purpose embodied AI through an end-to-end unified large model. It recently unveiled the Quanta X2, a fully self-developed wheeled dual-arm humanoid robot that seamlessly integrates with multimodal large model control. Independent Variable's concurrent software and hardware development strategy, coupled with its visionary technical concepts and accomplishments, has garnered endorsements from national investment platforms, premier domestic and international investment firms, and industrial capital.

As the pioneering Chinese company to achieve an end-to-end embodied AI large model, Independent Variable has independently crafted the "WALL-A" series of VLA (Vision-Language-Action) operational large models, establishing a unified cognitive and action framework. Within this unified representation space, the model concurrently processes perception, reasoning, and action, directly executing cross-modal causal reasoning and action decision-making, thereby enabling robots to ultimately think and operate akin to humans. Presently, the "WALL-A" model has exhibited zero-shot generalization capabilities in certain novel task types that have not undergone any training.

Concurrently, the company has introduced an innovative end-to-end embodied chain-of-thought reasoning framework. This framework conducts in-depth reasoning based on multimodal inputs and generates multimodal outputs, forming a complete autonomous loop of model decision-making, execution, exploration, and reflection. The model adeptly integrates language understanding, visual perception, and action execution, fostering a reasoning process that closely mirrors human cognition. It has successfully overcome the challenges of multi-step long-sequence tasks, substantially enhancing task completion rates and significantly expanding the robot's capacity to manage intricate real-world scenarios.

In the middle of this year, the company accomplished, for the first time, the control of a high-degree-of-freedom dexterous hand by an embodied AI large model for intricate operations. Prior to this, Independent Variable released a video demonstrating its proprietary large model controlling a high-degree-of-freedom dexterous hand to delicately handle and distribute playing cards, which are elastic and prone to deformation.

Currently, to advance the research and application of embodied AI large models, Independent Variable has made its developer-oriented embodied foundational model, "Wall-OSS," open-source, along with publicly releasing the relevant training code. This facilitates swift fine-tuning and practical deployment by developers worldwide on their own robotic platforms.

Wall-OSS boasts robust generalization and reasoning capabilities, outperforming other foundational models in extended operational tasks. As a multimodal foundation model, it also showcases exceptional causal reasoning, spatial comprehension, and reflective abilities.

Wall-OSS is an open-source embodied foundational model trained on extensive real-world data. Regarding model architecture, it innovatively features a "Shared Attention + Expert Feedforward Network (FFN)" design, enabling seamless transfer of VLM pre-training knowledge to the operational model. Language, vision, and action information are all embedded and processed within the same representation space. In terms of training methodology, it pioneers a "discrete-first, continuous-second, joint-third" multi-stage training paradigm, preserving the language and visual understanding capabilities of VLMs while endowing them with fine-grained action execution capabilities. Additionally, the unified cross-level chain-of-thought facilitates seamless forward mapping across abstract levels, enabling the model to effortlessly transition between high-level decision-making and low-level execution within a single differentiable framework.

In the realm of hardware, Independent Variable unveiled the Quanta X2, a fully self-developed wheeled dual-arm humanoid robot, in August of this year. Within a span of less than six months, the company has achieved full-stack self-development of the robot's entire body, high-degree-of-freedom dexterous hands, and exoskeleton teleoperation data collection equipment.

The Quanta X2 is a model-native general-purpose robot body designed not only to cater to the demands of model training and complex operational tasks but also to achieve comprehensive balance and optimization in core metrics such as load capacity, operational space, motion speed, and control precision.

The five-fingered dexterous hand of the Quanta X2 adopts a bionic structural design, with each hand boasting 20 degrees of freedom and the capability to perceive subtle pressure variations. Meanwhile, leveraging arm-hand integrated exoskeleton technology, Independent Variable has pioneered an industry-leading "humanoid robotic arm + high-degree-of-freedom dexterous hand" integrated whole-body teleoperation solution. The Quanta X2 not only collects high-quality data to feed back into model training but also deeply integrates with the self-developed model, enabling true deployment and application in real-world scenarios.

With the augmentation of its integrated software and hardware capabilities, Independent Variable's robots have already forged collaborations with leading service and industrial clients and have been deployed across multiple scenarios. Looking ahead, Independent Variable will also partner with clients to construct an open ecosystem centered around models and hardware, propelling the further advancement of embodied AI.