Firstly, what is Deep Tech as opposed to Tech or technology enabled? Sometimes Deep Tech is regarded as a science based startup, sometimes it is regarded as disruptive to the status quo, sometimes it is regarded just as slow and hard, capital intensive, with a long ROI horizon. Or as something that investors aren’t ready for yet. But the amount of money going into Deep Tech investing is increasing, and the pool of Deep Tech investors is increasing. One of the key points I made in a recent GIST Tech Connect Deep Tech panel is that most investors, including the most successful Tech investors are not able to invest seriously in Deep Tech startups because they lack the technical awareness and depth of commercialization experience specific to a Deep Tech startup. GIST or the Global Innovation in Science and Technology Network is the US State Department program to encourage and support global entrepreneurship.
In fact, if you do the research into the failure rates of some high profile Deep Tech startups, it seems that certain large funds have a much higher failure rate than others, so at best, their growth pathway is not compatible with Deep Tech startups. At worst, they are simply cherry picking some Deep Tech startups for their publicity value. Startups should always do their due diligence on investors and how they treat founders, particularly founders with similar startups.
Universities play a huge role in derisking, funding and commercializing Deep Tech startups but there is still a ‘Valley of Death’ in the transfer stages. And a Deep Tech startup can come out of any university but not all universities have real commercialization experience and a supportive startup ecosystem. Silicon Valley Robotics and Circuit Launch have provided a ‘halfway house’ for a lot of Deep Tech startups by providing affordable workspace with prototyping facilities and a startup ecosystem. But the first question I always ask entrepreneurs is if they have leveraged every advantage that their university connections can provide. Universities can provide greatly discounted lab space and testing facilities, also connections to scientific experts in most any field who can be leveraged as consultants and advisors.
The SBIR program, or the American Seed Fund, which is about a $4 billion non dilutive funding from the federal government in the form of R&D dollars, contracts and grants to small businesses and startups gives you the opportunity to derisk a lot of the technology very early on. You can really do a detailed scope and scan, and then couple that with the iCorps program and you get the opportunity to do deeper dives into customer discovery, to really understand if this is something that’s just a nice to have, or is it a real must have. Although the SBIR program is American based program, a lot of the countries around the world have been creating similar ones. A good example of that is EU Horizon 2020 grants.
Grants catalyze and do a certain amount to de-risk technology, extending the runway through non-dilutive funding and by creating a technology roadmap which validates the science as significant. Corporate venture funds or strategic investors also play an important role, alongside non-dilutive grant funding. Not only can they be a check, they can be a customer, they can be an advisor and a partner in the early prototype to manufacture stages. The best strategic investors play a huge role in helping Deep Tech startups succeed, because they need the technology you are creating.
Here’s a collection of tips for Deep Tech founders gathered from the GIST TechConnect Panel on Deep Tech with Nakia Melecio from Georgia Tech, Nhi Lê from WARF, Andra Keay from SVR and The Robotics Hub, G. Nagesh Rao from US Dept of State. Also tips from Six red flags that send investors running the other way by Sara Bloomberg, San Francisco Business Times. Quotes not attributed to other investors are my thoughts or recollections from the event.
“When you start going from accelerator to accelerator looking for funding, then you’re doing it wrong. Accelerators only fund you to participate in their program. Their program and mentors are the real value.” Nhi Lê, WARF Accelerator
You also dilute your equity and become uninvestable.
Always negotiate terms. But don’t focus solely on the financials and at the risk of throwing away the less obvious value that a good investor can bring to you.
“Deep Tech startups may take longer to get to revenue than a traditional tech startup, so you need to think about grant funding as a source of revenue, and any contracts that help you develop part of your technology.” Nhi Lê, WARF Accelerator
“Companies often say that they’re investable because they have a patent, but they haven’t budgeted anything to defend it. Your IP is only as good as your ability to defend it. Universities play a great role in protecting and defending IP that they’ve licensed.” Nhi Lê, WARF Accelerator
“When you go into your first funding meeting, you must be thinking about the long term journey, all the way to exit. It’s never going to be just one check, you’re growing a company.” G. Nagesh Rao, US Dept of Commerce
“Deep tech, especially at the leading edge, is usually expensive, so it’s critical to find the right path to commercialization at scale. Good investors speed up the process and lower your burn rate.” Michael Harries, The Robotics Hub
Have your potential investors brought similar startups to market? Having that experience can make the commercialization process much faster, and it’s critical to manage your resources effectively. Constant fundraising takes founders away from product development. Also, do your investors have patient capital? Or are they needing a rapid return on investment for their current fund? Don’t assume that a well known investor or accelerator guarantees you success, or even finding a good fit with their process.
Overreaching on inventory, being unable to meet debts in a timely fashion, structuring the company poorly, all these things are cited by founders who’ve struggled.
“Focus on the customer and fall in love with the customer’s problem and you’ll never go wrong.” Nakia Melecio, Georgia Tech
Do it from the start, and never stop going to market. You can’t just outsource your business development to people with better sales skills, not until you know that pain points you’re solving for your customers and you can write the scripts for them.
“If a founder is estimating their market in the trillions of dollars they have either not done the research or they are just delusional.” Swati Chaturvedi, Co-Founder of PropelX
“Founders who are focused only on vanity metrics (growth rate and valuation) and not attuned to developing sound business models are a red flag.” Anurag Chandra, Fort Ross Ventures
“Another red flag is trying to FOMO you into moving quickly. Not only is it bad for arriving at a sound investment decision, it’s an indication of how they do business with customers and partners (ie. not invested in building long term relationships). Anurag Chandra, Fort Ross Ventures
“Investors can tell when you are avoiding details like actual product or customer development status and it may mean you are misrepresenting your business.” Caroline Winnett, Executive Director of Berkeley SkyDeck
“There needs to be agreement on who is acting as CEO, and everyone needs to be aligned on that. Another red flag is not being open to advice from experts.” Caroline Winnett, Executive Director of Berkeley SkyDeck
“Founders should be responsive to requests for more information. It shows if they are organized and in the mindset to do a deal versus spin cycles.” Shruti Gandhi of Array Ventures
In this episode, Audrow Nash speaks to Youssef Benmokhtar, CEO of GelSight, a Boston-based company that makes high resolution tactile sensors for several industries. They talk about how GelSight’s tactile sensors work, GelSight’s new collaboration with Meta AI (formerly, Facebook AI) to manufacture a low cost touch sensor called DIGIT, on the digitization of touch, touch sensing in robotics, how GelSight is investing in community and open source software, and Youssef’s professional path in several industries.
Episode Links
Podcast info
Congratulations to Huy Ha and Shuran Song who have won the CoRL 2021 best system paper award!
Their work, FlingBot: the unreasonable effectiveness of dynamic manipulations for cloth unfolding, was highly praised by the judging committee. “To me, this paper constitutes the most impressive account of both simulated and real-world cloth manipulation to date.”, commented one of the reviewers.
Below, the authors tell us more about their work, the methodology, and what they are planning next.
In my most recent publication with my advisor, Professor Shuran Song, we studied the task of cloth unfolding. The goal of the task is to manipulate a cloth from a crumpled initial state to an unfolded state, which is equivalent to maximizing the coverage of the cloth on the workspace.
Historically, most robotic manipulation research topics, such as grasp planning, are concerned with rigid objects, which have only 6 degrees of freedom since their geometry does not change. This allows one to apply the typical state estimation – task & motion planning pipeline in robotics. In contrast, deformable objects could bend and stretch in arbitrary directions, leading to infinite degrees of freedom. It’s unclear what the state of the cloth should even be. In addition, deformable objects such as clothes could experience severe self occlusion – given a crumpled piece of cloth, it’s difficult to identify whether it’s a shirt, jacket, or pair of pants. Therefore, cloth unfolding is a typical first step of cloth manipulation pipelines, since it reveals key features of the cloth for downstream perception and manipulation.
Despite the abundance of sophisticated methods for cloth unfolding over the years, they typically only address the easy case (where the cloth already starts off mostly unfolded) or take upwards of a hundred steps for challenging cases. These prior works all use single arm quasi-static actions, such as pick and place, which is slow and limited by the physical reach range of the system.
In our daily lives, humans typically use both hands to manipulate cloths, and with as little as a single high velocity fling or two, we can unfold an initially crumpled cloth. Based on this observation, our key idea is simple: Use dual arm dynamic actions for cloth unfolding.
FlingBot is a self-supervised framework for cloth unfolding which uses a pick, stretch, and fling primitive for a dual-arm setup from visual observations. There are three key components to our approach. First is the decision to use a high velocity dynamic action. By relying on cloths’ mass combined with a high-velocity throw to do most of its work, a dynamic flinging policy can unfold cloths much more efficiently than a quasi-static policy. Second is a dual-arm grasp parameterization which makes satisfying collision safety constraints easy. By treating a dual-arm grasp not as two points but as a line with a rotation and length, we can directly constrain the rotation and length of the line to ensure arms do not cross over each other and do not try to grasp too close to each other. Third is our choice of using Spatial Action Maps, which learns translational, rotational, and scale equivariant value maps, and allows for sample efficient learning.
We found that dynamic actions have three desirable properties over quasi-static actions for the task of cloth unfolding. First, they are efficient – FlingBot achieves over 80% coverage within 3 actions on novel cloths. Second, they are generalizable – trained on only square cloths, FlingBot also generalizes to T-shirts. Third, they expand the system’s effective reach range – even when FlingBot can’t fully lift or stretch a cloth larger than the system’s physical reach range, it’s able to use high velocity flings to unfold the cloth.
After training and evaluating our model in simulation, we deployed and finetuned our model on a real world dual-arm system, which achieves above 80% coverage for all cloth categories. Meanwhile, the quasi-static pick & place baseline was only able to achieve around 40% coverage.
Although we motivated cloth unfolding as a precursor for downstream modules such as cloth state estimation, unfolding could also benefit from state estimation. For instance, if the system is confident it has identified the shoulders of the shirt in its state estimation, the unfolding policy could directly grasp the shoulders and unfold the shirt in one step. Based on this observation, we are currently working on a cloth unfolding and state estimation approach which can learn in a self-supervised manner in the real world.
|
Huy Ha is a Ph.D. student in Computer Science at Columbia University. He is advised by Professor Shuran Song and is a member of the Columbia Artificial Intelligence and Robotics (CAIR) lab. |
|
Shuran Song is an assistant professor in computer science department at Columbia University, where she directs the Columbia Artificial Intelligence and Robotics (CAIR) Lab. Her research focuses on computer vision and robotics. She’s interested in developing algorithms that enable intelligent systems to learn from their interactions with the physical world, and autonomously acquire the perception and manipulation skills necessary to execute complex tasks and assist people. |