It is a widely known fact, that the last mile (or kilometre) delivery before the parcel/package reaches the end destination causes a large portion of the total transportation costs. There are many reasons for that, but the core reason is naturally the economics of scale. Transporting thousands/millions parcels with an ocean-going ship is so much more cost-effective than to have a single parcel transported from a local post office to a person living in an apartment on the fifth floor.

To reduce that high cost per delivered parcel, the study for autonomous last-mile delivery robots has been extremely active for, at least, the past 20 years. There are two main paths: the airborne and the land-based solutions. As we have stayed out of the drone business, our focus has been on the land-based solutions. Although there are some great solutions using modern walking machines for the task, the majority rely on wheel-based solutions.

Back in the days

GIM Robotics has been doing logistics solutions from the day one. Actually, we have been doing that for 30 years, if we count our days at one of the renowned mobile robotics laboratories in the world since late 1980’s, in Helsinki University of Technology, later Aalto University. We did these solutions during the times when GPS signal was jumping like a crazy rabbit all over the place. We spent good money on the top-of-the-line IMUs, external supportive solutions and later turned novel industrial 2D LiDARs into 3D LiDARs with some homemade mechatronic solutions. Probabilistic robotics became our religion early on and we have not looked back ever since. When RTK-GNSS became widely available, reasonable priced good-enough IMUs were all over the place and ready-made 3D LiDARs were brought to the markets, things got so much easier.

From university to industry

After University days and particularly after our six years long Academy of Finland funded Centre of Excellence period, we were ready to jump on the other side of the fence and started GIM Robotics in 2014. Based on our inhouse analysis of our focus areas, various logistics related projects, PoCs and feasibility studies have been our bread and butter already for many years. 

Public projects

In most of the cases, we have retrofitted our clients’ machines, or tuned commercially available robotic platforms for the job. Unfortunately, those projects are protected by NDAs. However, for several years, we have been manufacturing our own logistics platforms suitable for both indoor and outdoor operations. First version, SpOv1, has been successfully operating in Paris, France on Nokia Bell Labs campus (Left video, below) in EIT-Digital funded LMAD project (now LMAD, the company) and the second updated version, SpOv2, has been performing pilots in various location in Otaniemi, Espoo (Middle) and Helsinki (Right) together with various logistics companies including DB Schenker.

The workhorse

The newest version, dubbed surprisingly as SpOv3, was built solely for our own usage and we have used it for various perception system testing and full-stack optimizations in various public projects like MURO and PEAMS. Basically, we wanted to have the newest and the best platform for our extensive R&D operations and especially for related position and mapping SW stack improvements and long-term validations. We wanted to demonstrate that our solutions survive in all conditions, because we forecast that many companies will expand their logistics solutions to serve both indoors and outdoors environments – there are plenty of economically justifiable single machine solutions serving both indoor and outdoor areas.

More is better

Our fundamental philosophy when designing solutions for our client’s logistical problems, has been to initially test the proposed machines (new or retrofitted ones) with an extensive sensory suite. That approach gives us a clear view, which perception sensors or combinations are vital for optimal (or near-optimal) performance in any particular use case including important ODD (Operational Design Domain) definitions and restrictions. We have deep knowledge of most of the imaginable sensors, including LiDARs, radars, various camera solutions etc. Naturally we integrate the usual suspects, like IMUs, RTK-GNSS and odometry info, to our standard solutions.

Getting the data

To get the extensive amount of data needed for algorithmic development and SW quality validation for our main offerings, we needed to come up with a challenging use case, which would offer us an easy way to benefit from our own SpOv3 robot and to have extensive and frequent data sets from a difficult environment including some traditional elements normally related either to indoor or outdoor environments. In our case it means, that we will have our robot traveling frequently through some GNSS-challenging and -denied areas. We were happy to notice that the route we have been walking during our lunch breaks from our HQ to nearby large shopping mall includes several long tunnels. Furthermore, it is inherently a dynamic environment due to many pedestrians and cyclist and frequent impromptu road works. It is hard to imagine better route for our purposes.

Proven solutions

Just like with any other autonomous mobile robot application, the last mile delivery case requires localization, situational awareness and task/mission performing capabilities. PAIKKA localizes the machine in any weather conditions with centimetre-level accuracy. KARTTA creates and updates the 3D map of the environment and TARKKA provides the needed real-time situational awareness around the robot. All those modules have been tested extensively, in this context and beyond.