Danu's retrofit recycling robot counts a pick only once the item reaches its intended bin

Counting at the bin confirms that the arm delivered the item the vision system selected ^[6][19]. It says less about the selection. If the camera misreads a piece of plastic, that piece still lands in its intended bin and still counts ^[19]. Runtimewire's report makes the same point: the count does not by itself establish how accurately H.E.R.O. sorts at commercial scale ^[20]. A sales deck would not need a definition this strict. I think it is the right one, because the reported figure is tied to material that actually arrived in a bin ^[19].

Ma came to the problem as a software developer at a London bank ^[3]. "All these different recycling bins that just end up in the same place," she told TechCrunch ^[4]. Downstream of the bins, she found plants that still relied heavily on people separating mixed waste by hand ^[5]. She founded Danu in Edinburgh, and the sorter took six years of development to reach market ^[5][1]. A robot in that setting has to fit a facility built for people and handle inconsistent material. It also has to be worth the money to an operator whose revenue rests on the material it can recover and sell, runtimewire notes ^[12]. The sub-metre footprint and sub-day install deal with the first of those constraints ^[17]. Danu says grippers can be swapped in 20 minutes ^[18]. I also like the decision to license the software apart from the custom hardware ^[8]. Danu's pitch is that its software and AI improve as the system processes more material and gets more vision data ^[13]. The separate license exists so that software can be updated as the technology develops ^[8]. The report does not say whether Ma's $24,000 annual maintenance figure includes the license ^[10].

Taken at face value, Ma's estimate nets $301,000 in the first year after the $160,000 outlay and $24,000 of maintenance ^[21]. On revenue net of maintenance, the outlay pays back in about four months ^[22]. That figure is revenue. Runtimewire stresses that it is Ma's estimate of added revenue, not verified customer savings or profit. It adds that its value depends on which materials the machine recovers consistently, the site's volume and the cost of running the whole sorting operation ^[11]. For a four-month payback to carry over to another plant, that plant would need a similar material mix and volume. The extra recovered material would also need to cost little to process and sell.

The contracts are the other half of the evidence. At Ma's $160,000 initial-investment figure, $500,000 would cover about three sites if every contract dollar went to hardware ^[23]. It is a tenth of the $5 million late-seed round ^[24][16]. Runtimewire ranks the sales stages. Signed contracts carry more weight than letters of interest, and prospects do not establish orders, deployments or recurring revenue ^[14]. TechCrunch puts the sorting opportunity across Europe and North America at $20 billion ^[15].