You upload a video of a person picking a strawberry. The system watches the bend of the wrist, the pinch of the fingers, the careful twist. Later, a robot arm in a field replicates the motion. This is the foundational promise of Nupeak, a company in Burnaby that builds what it calls "AI-driven industrial robotics with human-centric training" [nupeak.com]. It is a quiet bet, one that lives more in the logic of its demo than in a flood of press releases. The company's website shows two products: Pixa, a harvesting robot claiming a 90%+ pick rate, and Hive, the software platform that turns those training videos into executable actions [nupeak.com/hive]. The interface is clean, the typography modern, the promise direct. It feels like a product built for an operator who needs a solution, not a venture capitalist who needs a story.
The Wedge of Human Demonstration
Nupeak's core technical bet is that the most effective way to teach a robot complex, delicate tasks is not through exhaustive code but through demonstration. The Hive platform is positioned as an intelligent training manager, handling the full pipeline from data collection to model deployment and fleet management [nupeak.com/hive]. This approach is a specific answer to a specific problem in physical AI: variability. No two berries, no two branches, no two packing lines are identical. By training on human video, the system ostensibly learns not just a single motion but an adaptable policy for handling real-world messiness. It is a classic robotics wedge, using software to make hardware smarter and more flexible, starting with the acute labor pains of agriculture.
A Stealthy Path to the Field
The company's public footprint is notably light. Founded in 2019, it has raised an undisclosed amount across six funding rounds from eleven investors, including a $100,000 grant in late 2020 [Crunchbase]. The leadership team is headed by co-founder and CEO Anshul Porwal, an engineering physics graduate from the University of British Columbia with a background in UAVs and sensor design [Crunchbase][LinkedIn]. CTO Div Gill rounds out the known technical leadership [LinkedIn]. The company lists offices in Burnaby and Victor, New York, and public records suggest a headcount between 9 and 50 employees [Apollo][ZoomInfo]. This relative stealth extends to customer traction. While third-party sites note deployments of "Pixaberry" robots and partnerships with farms like Bergen Farms and Hilliers Estate Farms are mentioned in regional reporting, these are not prominently featured on Nupeak's own site, leaving the scale of commercial deployment unclear [Techcouver][HiHello].
| Role | Name | Background Note |
|---|---|---|
| Co-Founder & CEO | Anshul Porwal | Engineering Physics, UBC; 5+ years in UAVs & sensor design [Crunchbase][LinkedIn] |
| CTO | Div Gill | Technical leadership role [LinkedIn][Apollo] |
The Risks of a Quiet Build
For all the elegance of its premise, Nupeak operates in a space where ambition is often ground down by physics, cost, and scale. The company's quiet profile presents a dual-edged risk. On one hand, it suggests a focus on product and early pilot customers over hype. On the other, it raises questions about velocity, commercial validation, and competitive visibility. The agricultural robotics sector is crowded with well-funded players tackling harvesting, weeding, and scouting. Nupeak's differentiation rests on its Hive training platform,a software layer for continuous learning. Yet, without public metrics on fleet size, pick-rate validations from third parties, or detailed customer case studies, it is difficult to assess how that differentiation translates into a durable advantage. The company's next phase will likely hinge on moving from promising pilots to repeatable, scaled deployments that prove both the unit economics and the reliability of its human-trained robots.
Ultimately, Nupeak is answering a cultural question as much as a technical one. It asks what happens when we stop trying to program robots to think like machines and start teaching them to act like us. The product implies that the intelligence needed for the nuanced work of harvesting isn't found in a pure simulation or a massive dataset of labeled images, but in the tacit knowledge of a skilled worker's hands. The success of the bet won't be measured in model parameters, but in whether a grower, watching a robot mimic a picker's deft movement, sees a replacement or a revelation.
Sources
- [nupeak.com] Nupeak - AI driven industrial robotics with human centric training | https://nupeak.com/
- [nupeak.com] Hive: Intelligent Robot Training Platform - Nupeak | https://nupeak.com/hive
- [Crunchbase] Anshul Porwal - Crunchbase Person Profile | https://www.crunchbase.com/person/anshul-porwal-c2d3
- [LinkedIn] Anshul Porwal - CEO & Co-Founder at Nupeak Robotics | https://ca.linkedin.com/in/cybrfish
- [Crunchbase] Nupeak - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/nupeak
- [Crunchbase] Grant - NeuPeak Robotics - 2020-10-13 - Crunchbase Funding Round Profile | https://www.crunchbase.com/funding_round/neupeak-robotics-grant--81f34766
- [Apollo] Company profile data for Nupeak | Source not linked
- [ZoomInfo] Neupeak Robotics - Overview, News & Similar companies | https://www.zoominfo.com/c/neupeak-robotics/479489709
- [Techcouver] Article referencing Nupeak partnerships | Source not linked
- [HiHello] Pixaberry robots deployed in agriculture | http://hihello.ca/about.html