MirrorSpace Raises $10M Seed for Robot Spatial Intelligence

A nascent startup, MirrorSpace, is making a significant play in the burgeoning field of embodied AI by developing advanced spatial perception systems for robots. The company, founded by a trio of PhDs averaging just 24 years old, has successfully closed a ten million USD seed funding round. This capital infusion was led by prominent venture firms Songhe Capital and Matrix Partners China, signaling strong investor confidence in the team's vision.

MirrorSpace aims to bridge a critical gap in the robotics industry: enabling robots to not only move with agility but also to truly understand and interact with their physical environments. While current robots excel at motion control, their ability to perform complex tasks is often hampered by a lack of sophisticated spatial awareness. The company is focused on creating a foundational "spatial brain" for robots, moving beyond simple sensor data processing to offer comprehensive environmental comprehension.

The core innovation from MirrorSpace lies in its proprietary perception module, codenamed MirrorSense. This system integrates not only traditional visual and LiDAR data but also thermal information, creating a richer, more robust spatial representation. By performing deep, early fusion of heterogeneous sensor inputs at the edge, MirrorSense achieves lossless spatio-temporal alignment of RGB, depth, and thermal data. This capability is crucial for enabling reliable robot operation in challenging conditions such as low light, glare, occlusions, and environments with low texture detail.

Furthermore, MirrorSpace is differentiating itself by incorporating "time" as a fourth dimension in its spatial modeling, moving beyond static 3D reconstructions. This dynamic spatio-temporal perception stream addresses the limitations of traditional methods in capturing the nuances of moving environments. The company's Mirror-Mind decision-making hub then aligns this fused 4D representation with large vision-language models (VLMs), enabling robots to build a persistent, queryable spatial memory. This allows for more intelligent decision-making, such as a robot recalling the location of a previously seen object even if it's currently obscured.

The founding team, including CEO Jin Yili (a McGill University PhD with expertise in spatial intelligence and multimedia systems), and co-founders Hu Kaiyuan and Duan Xize, brings a strong academic pedigree. Their advisor, Liu Xuewei, a Fellow of the Canadian Academy of Engineering and IEEE, adds significant depth with his extensive experience in spatial intelligence and machine learning. This blend of academic rigor and practical application focus impressed investors, with Songhe Capital highlighting the team's "judgment beyond their age" and Matrix Partners China noting their "complete cognitive chain from academia to industry" and early signs of "real willingness to pay" in industrial settings.

MirrorSpace is pursuing a dual business strategy. The company plans to directly supply its integrated spatial perception solutions to robot manufacturers, enhancing or replacing existing sensor modules. Concurrently, they are collaborating with system integrators and platform providers for deployments in sectors like industrial inspection (power, petrochemicals) and commercial applications including high-end real estate, retail, and the entertainment industry. The company anticipates achieving over ten million USD in orders by 2026, with its prototype module already garnering significant customer interest.

Looking ahead, MirrorSpace envisions its MirrorSense modules becoming a standard infrastructure for physical world interaction for all robots. By accumulating vast amounts of real-world spatio-temporal data through scaled deployments, the company aims to provide foundational datasets for training next-generation world models. While acknowledging the challenges of scaling engineering and supply chain operations, the young team is confident in its ability to carve out a niche in a competitive market, citing their strong algorithmic capabilities as a key differentiator against established players like Orbbec and Hikvision, as well as in-house solutions from robot makers like Unitree Robotics.