Covariant is an artificial intelligence robotics company that develops software for industrial robotic arms used in warehouse automation. Founded in 2017 by former OpenAI researchers, the company focuses on creating general-purpose AI systems that enable robots to perceive, reason, and act in unstructured environments, particularly for tasks like item picking, packing, and sorting in logistics centers. Its core product, the Covariant Brain, is a cloud-connected AI platform that learns from real-world data across multiple deployments.
The company emerged from research at the intersection of deep learning and robotics, aiming to move beyond traditional pre-programmed automation. Covariant's approach leverages reinforcement learning and neural networks to allow robots to handle novel objects and adapt to changing conditions, a significant departure from conventional industrial robots that require rigid, task-specific programming. The company has positioned itself as a key player in the growing field of physical AI, where machine learning models are applied to real-world manipulation challenges.
Founding and Early Development
Covariant was founded in 2017 by Pieter Abbeel, Peter Chen, and Rocky Duan, all of whom had backgrounds in artificial intelligence research. Abbeel, a professor at the University of California, Berkeley, had previously worked on deep reinforcement learning and robotics, while Chen and Duan were early members of OpenAI, contributing to research on generative models and reinforcement learning. The trio sought to apply cutting-edge machine learning techniques to industrial robotics, a sector that had largely relied on deterministic control systems.
The company initially operated in stealth mode, developing its technology with funding from prominent venture capital firms. In 2018, Covariant raised $20 million in a Series A round led by Amplify Partners, with participation from other investors. By 2019, the company had begun pilot deployments with logistics partners, testing its AI-driven robotic picking systems in real warehouse environments. These early trials demonstrated the ability of the Covariant Brain to handle a wide variety of items, including those with irregular shapes, reflective surfaces, and varying degrees of fragility.
Technology and Products
The Covariant Brain is the company's flagship product, a software platform that integrates with third-party robotic arms, grippers, and sensors. The system uses deep learning models, including convolutional neural networks for vision and reinforcement learning for control, to perform tasks such as grasping, placing, and sorting. Unlike traditional vision-guided robotics that rely on explicit geometric models, Covariant's approach learns from data collected across its fleet of deployed robots, allowing the system to improve over time.
A key feature of the platform is its cloud-based architecture. Robots in different warehouses send anonymized data to Covariant's servers, where models are retrained and updated, then deployed back to the fleet. This continuous learning loop enables the system to generalize to new objects and scenarios without requiring on-site reprogramming. The company also offers a simulation environment for testing and validation, which reduces the need for physical trials.
In 2021, Covariant introduced the Covariant Brain for Picking, a turnkey solution for piece-picking applications in e-commerce fulfillment centers. This product integrates with major robotic arm manufacturers, such as ABB and Fanuc, and includes a suite of software tools for order picking, induction, and kitting. The system is designed to operate at human-level speed and accuracy, with reported pick rates of up to 1,000 items per hour in some deployments.
Commercial Deployment and Growth
Covariant's technology has been deployed in warehouses operated by major logistics and retail companies. In 2020, the company announced a partnership with Knapp, an Austrian automation provider, to integrate the Covariant Brain into Knapp's sortation systems. This collaboration expanded Covariant's reach into European markets. By 2022, Covariant had raised over $220 million in total funding, including a $75 million Series C round led by Index Ventures, with participation from existing investors and new backers like Temasek.
The company has also expanded its product line to include the Covariant Sortation System, which uses robotic arms to sort items into different chutes or bins based on barcode or visual recognition. This system is particularly useful for handling irregularly shaped items that are difficult for traditional conveyor-based sorters. In 2023, Covariant reported that its robots had performed over 100 million picks across its customer base, a milestone that underscored the scalability of its AI approach.
Research and Industry Position
Covariant maintains a strong research orientation, publishing papers on topics such as reinforcement learning for robotic manipulation and sim-to-real transfer. The company's founders and researchers have contributed to academic conferences, including the Conference on Robot Learning and the International Conference on Robotics and Automation. This research focus distinguishes Covariant from more traditional automation vendors, positioning it as a bridge between academic AI and industrial application.
The company operates in a competitive landscape that includes other AI robotics startups like Figure AI and Sanctuary AI, as well as established players like Intuitive Surgical in medical robotics. However, Covariant's niche in warehouse logistics is distinct, with direct competitors including companies like Fermata (which focuses on agricultural robotics) and Braina (a different AI venture). Covariant's emphasis on general-purpose learning rather than task-specific programming is seen as a key differentiator, aligning with broader trends in Generative AI and Large language model research, though its models are specialized for physical actions rather than text.
Future Directions and Challenges
As of 2024, Covariant continues to expand its deployment footprint and refine its AI models. The company has explored applications beyond warehousing, including parcel sorting for postal services and depalletizing in manufacturing settings. However, the broader adoption of AI robotics faces challenges, including high upfront costs, integration complexity, and the need for reliable performance in safety-critical environments. Covariant addresses these by offering software-as-a-service pricing models and emphasizing the flexibility of its learning-based approach.
The company's long-term vision is to create a "universal AI for robots" that can be applied across diverse industries, from agriculture to healthcare. This ambition aligns with the work of other research groups, such as BAIR (Berkeley AI Research) and Stanford AI Lab, which are exploring similar foundational models for robotics. Covariant's success will depend on its ability to maintain its technological lead while scaling commercially, a challenge that has proven difficult for many AI startups in the physical world.
See Also
- Artificial intelligence
- Machine learning
- Reinforcement learning
- Robotics
- warehouse-automation
References
- Covariant official website, company history and product information.
- TechCrunch, "Covariant raises $75M for AI-powered warehouse robots," 2022.
- IEEE Spectrum, "Covariant's AI robots learn to pick anything," 2021.
- Company press releases and funding announcements, 2018-2023.