Team at Covariant

Covariant is building universal AI that allows robots to see, reason, and act on the world around them. We are bringing this technology to commercial viability, starting with the industries that make, move and store things in the physical world.

Job Openings at Covariant

Top Engineering Values

Each team is asked to select, explain, and rank their top 8 values in order of importance.
  • Creative + Innovative

    We bring cutting-edge AI research and world-class engineering into the real world.

    Since the first industrial robots were introduced in the 1960s, millions have been deployed globally. Their impact is undeniable. Robots have automated countless dangerous, repetitive tasks, transforming manufacturing, but they've only reached a fraction of their potential. Incapable of thinking on their own, they can only do pre-programmed tasks in tightly-controlled environments. They can't understand, learn, or adapt. Covariant was founded in 2017 to change this.

    Building on our experience at Berkeley and OpenAI, our vision is the Covariant Brain: universal AI that allows robots to see, reason, and act on the world around them. We’re bringing the Covariant Brain to commercial viability, starting with the industries that make, move, and store things in the physical world.

    Our approach is rooted in the following principles:

    1. General robotic intelligence. Instead of learning to master specific tasks separately, Covariant robots learn general abilities such as robust 3D perception, physical affordances of objects, few-shot learning, and real-time motion planning. This allows them to adapt to new tasks just like people do – by breaking down complex tasks into simple steps and applying general skills to complete them.
    2. Research meets industry. Bringing practical AI Robotics into the physical world is hard. It involves giving robots a level of autonomy that requires breakthroughs in AI research. That’s why we have assembled a team that has published cutting-edge research papers at the top AI conferences and journals, with more than 50,000 collective citations. In addition to our research, we’ve also brought together a world-class engineering team to create new types of highly robust, reliable, and performant cyber-physical systems.
    3. Perfecting each component. We’re only as strong as our weakest link. If one component lags, the whole system fails. That’s why we’ve built a full-stack team, investing in making each robotic component world-class, from software systems to hardware stations, from AI algorithms to end-effectors.
  • Safe Environment to Fail

    “Taking on the impossible, together” is one of our core values.

    Your current expertise does not limit what you are able to do or work on here at Covariant. Many of us have research backgrounds in academia, where failure is common and expected. We view it as a strength that our domain knowledge differs so widely across the company, which is why we encourage people to speak up and ask questions if they don’t know or understand something. You won’t know everything (no one can!), but embracing this truth gives us the confidence to make risky bets and try new things. Not only does this help us to grow, but this is how we as a company and industry innovate.

    Because robotics is so cross-functional, when something isn’t working, it takes an open mind to find the root cause. Triaging each problem means considering mechanical, electrical, general software, or AI errors. Some organizations get stuck blaming the weakest link. At Covariant, we are hyper-focused on making sure our customers are absolutely thrilled with our products and to do so effectively means short-circuiting that failure mode of blame. Success for us requires teamwork so we fully embrace every challenge as a single unit.

  • Open Communication

    We strive for empathy.

    We value EQ > IQ and are humble, considerate, and tight-knit. Our founders talk to us regularly; they’re honest and transparent about how the business is running and there are direct lines of communication between every employee and our leadership team. It’s important that we foster an environment where every team member feels comfortable talking and empathy is key. As a growing team (currently we have ~40 employees), things are always a work in progress. Individuals often speak up and surface concerns or criticism when they want to, and we all listen.

    During our most recent offsite, our entire company gathered in a circle one evening to give kudos to one another. The prompt was simply, “Feel free to thank someone who you’d like to recognize.” What started as an informal exercise turned into a two-hour event. The amount of transparency made us feel closer to one another and helped us learn what each member of our team really cares about and values. (We suspect giving kudos will become a tradition of ours!)

    Lastly, we practice open communication every day at the office. The entire team currently works in one large warehouse. This facilitates a tremendous amount of conversation and provides exposure to all aspects of the company. During lunch, you might hear about which partnerships our business development team is working on or what optimizations our hardware engineers are focusing on. Having an open office space (and still being a small team) enables us to exchange ideas and collaborate easily. We have ambitious goals, and we can only achieve them by supporting and working with each other!

  • Cross-Department Collaboration

    It’s a fundamental fact that robotics is interdisciplinary.

    We have a flat structure at Covariant, so we don't have rigid departmental divides. Everything we do is cross-functional because real progress in robotics requires tight integration of hardware, software, research, and business. For example, as Carlos Florensa (a PhD candidate) explained, “My research background hadn’t trained me to efficiently contribute to a constantly-changing codebase – I used to be the only person modifying my code! Thanks to the extremely collaborative nature of Covariant, and the patience and eagerness to help from colleagues like Jasmine Deng and co-founder Rocky Duan, I could ramp up these skills faster than anywhere else!”

    At Covariant we have all types of engineers, including data, mechanical, software, robotics, and machine learning engineers, just to name a few. Given our vast range of area expertise, it’s imperative that we are open to collaborating with one another on a regular basis. This goes beyond just engineering – you’ll also interact with people working on business development, web design, and marketing. And regardless of what you do, you will be within 40 feet of a robot. We are all self-starters who like bridging our knowledge gaps and working together with folks to make the most impact.

  • Customer Comes First

    Addressing real needs is what takes us from research to real value.

    Our founders Peter Chen, Rocky Duan, Tianhao Zhang, and Pieter Abbeel have published some of the most highly-regarded research in AI Robotics at the Robot Learning Lab at UC Berkeley. However, they were dissatisfied that their research wasn’t materializing into real world solutions. As a result, they came together to start Covariant and focus on existing problems and serving real people who face them.

    What sets Covariant apart from other AI and robotics companies is that we don’t search for problems that fit our state-of-the-art research. Instead, we start with our customers. There are many existing, real-life problems and our approach always begins with understanding what issues our customers are already facing.

    Before building anything, our team actually went out and spoke to customers to narrow ideas down to a targeted vertical: robotics that help in logistics and fulfillment.

    For example, our software directly impacts big logistics companies and online stores by helping them fulfill orders in a faster, more efficient way. Focusing on what our customer needs and regularly checking in to address their problems is a strategy that keeps us grounded, not in an ivory tower. And unlike other AI companies with hopes to solve real world problems some day, we started Covariant with actual customers. We have more demand than we can fulfill at this point, and stay aligned as a multidisciplinary team by focusing on serving our customers.

  • Impressive Team Members

    We’re proud of our collective knowledge, and impressed by the humility of our colleagues.

    We bring together domain knowledge across multiple fields and are truly innovating at the bleeding edge of technology. You feel honored to be sitting to the person next to you and you’re glad they’re on your team.

    Above all, there is a tremendous amount of humility on our team. Everyone is so approachable and happy to sit down with you to explain something or help you debug some code. Even our CEO and co-founder, Peter Chen (who was an instructor at UC Berkeley) enthusiastically hosts Q&As and leads reading groups to promote knowledge sharing. It is wonderful to work with smart people, but a completely different thing when they are the kind of people who are eager and willing to help up-level everyone around then.

  • Committed to Personal Growth

    The question is not if you will grow at Covariant, but how you will grow.

    We're tackling a problem that has no known solutions and no best practices, which means it isn’t even possible for us to hire someone who already has all of the necessary knowledge. Therefore, we have to embrace the idea that our very success as a business depends on our ability to learn and grow, both as individuals and as a team.

    At Covariant, it’s not about how much you know, but rather how much you are willing to learn. We all strive to become better versions of ourselves every day, and more importantly, we all see it as our responsibility to help our colleagues do the same. Domain experts host discussions and organize reading groups around topics they have a deep understanding in – from machine learning to goal setting – in order to help others expand their skill sets. Maximilian Sieb started a discussion group to get people up to speed in the basic methodologies related to computer vision (and more advanced topics that specifically relate to what we do at Covariant) and reported “how exciting it is to see people with no prior background in computer vision come up with innovative ideas only a few meetings in!”

    We help each other break out of our comfort zones and encourage one another to take on tasks we're interested in, even when they fall outside of our wheelhouse. Our strong internal support network and our desire to uplevel the people around us fosters an atmosphere where everyone feels encouraged to take on new tasks without ever being siloed.

  • Start-to-Finish Ownership

    See what you build in the hands of our customers, and get user feedback immediately.

    We have a tight feedback loop between makers and end users. What you design, code, and/or build is yours from beginning to end. Our customers are warehouse operators who can’t easily pop on or off updates or new models we deploy. They have to wait for a maintenance window, which means we may only have 30 minutes to drop in an update and make the decision whether to roll back or not. From early design stages to being on-call for production code, we are responsible for the success of our work through and through.

    Our sense of ownership is amplified because of how closely we monitor the performance of our deployed systems. We know our solutions are responsible for time sensitive shipments, and downtime is not acceptable. We live stream analytics to track regressions and proactively identify problems before they can cause downtime. To catch unexpected issues, we keep an open dialogue with our customers who can surface questions and alert us when the unforeseen arises. By having tight feedback loops, we never lose sight of who we’re building for and get the deep satisfaction of seeing our work being used by real people in the wild.

    You might even end up trying something you never envisioned. For example, Josh Mouledoux, a Mechanical Engineer, became our resident camera expert. As lead software engineer Andrew Vaziri can attest, “When I first joined Covariant, I set up the project management suite, which didn’t exist before. Now, I also run the patent process and help people communicate with lawyers. It’s something I never formally imagined when I was interviewing.” If you’re passionate about something, there’s room for you to put on another hat and try it.


  • Creative + Innovative
  • Safe Environment to Fail
  • Open Communication
  • Cross-Department Collaboration
  • Customer Comes First
  • Impressive Team Members
  • Committed to Personal Growth
  • Start-to-Finish Ownership

Company Properties

  • B2B
  • Technical Founder(s)

Team Members

  • 5 Business & Ops Team Members
  • 6 Founders & Leadership
  • 4 Hardware Engineers
  • 9 ML Researchers
  • 1 Product Manager
  • 1 Program Manager
  • 11 Software Engineers
  • 1 Solutions Engineer

Vacation Policy

Unlimited Vacation

Tech Stack

We believe in choosing technologies pragmatically, balancing tradeoffs between widespread support, power, and accessibility. Our primary language is Python, augmented with C/C++, PyTorch, and other extensions. We use code tools like Bazel, mypy, and automatic formatters to reduce the cognitive overhead of our work, and ops tools like Docker and Salt to automate operational burdens. Our code runs in a variety of environments, including AWS, Google Cloud, professionally-hosted private clouds, and most importantly, on our own servers at customer sites. We collaborate and document our work with Slack, GitHub, Jira, and Confluence.

Interview Process

We start with a phone call to share context and align expectations, followed by a take-home exercise or remote coding challenge. The final step is an onsite, which consists of roughly five panel interviews. Our interview process is designed to have a more “hands-on” feel: it covers a broad range of practical technical and non-technical skills that we believe are important, rather than strictly emphasizing algorithmic challenges.