PsiBot

The firm's core technology is centered on a hierarchical end-to-end architecture, often referred to as a "fast/slow brain" system, which combines intuitive motor control with deliberate reasoning and planning. This approach utilizes reinforcement learning (RL) to train robots, enabling them to learn and execute long-horizon, complex tasks that go beyond simple pick-and-place operations. PsiBot has introduced a series of end-to-end RL-based embodied models—Psi R0, R0.5, and R1—which have demonstrated capabilities such as dexterous manipulation, generalization to unseen objects, and long-term task reasoning in open environments. To facilitate this, the company has also developed a suite of supporting software and hardware, including the Psi Sim high-fidelity training platform, the Psi Data collection platform, and hardware like the Psi H1 dexterous hand and Psi E1 exoskeleton for data capture.

PsiBot's business model is B2B, targeting high-value scenarios in sectors like advanced manufacturing, retail logistics, and B2B services. The company's initial commercial applications focus on semi-structured environments such as barcode-scanning operations in clothing warehouses. PsiBot has secured significant funding to accelerate its growth, including an angel round led by GL Ventures and Lanchi Ventures, and a combined angel and Pre-Series A financing of approximately $280 million from investors including state-backed funds like Xuhui Capital. The capital is being used to expand deployments in logistics and build out a large-scale embodied AI data infrastructure, with a goal of accumulating one million hours of real-world training data by 2026.