Automation is imminent. More and more daily tasks are being conducted by machines/computers in many disciplines where a certain autonomy is being allowed to the machine, yet still in many cases hampered by regulations. Read more what the greatest challenges for forest robots are.
“It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.”
Why do we need robots?
Robots should assist humans when performing tasks that people would likely not do because of several reasons:
- Boring and repetitive tasks like work on assembly line
- Expensive and ineffective, when benefit is not big enough compared to the cost
- Dangerous and life threatening.
There are many reasons but the ones above are the most important arguments for using a robot.
Robots are getting more used in different industries but also in our households. Without getting into discussion on what can be defined as a robot, I will refer to it as all the machines with a certain degree of autonomy. In this context, an autonomous vehicle is a robot like a vacuum cleaner that drives autonomously in the house or a self-driven robo-mower. The point is, to reduce the human workload by use of the machines and at some point limit human work only to supervision.
Before starting to talk about robots in the forest, we should ask the question if there is a need to use them. It seems, that it makes sense to send out a “lumbeRobo” to cut the trees, and plant them, why wouldn’t we?
Is our technology advanced enough to introduce autonomy in forest operations? I will leave this question without an answer for now.
Forest robots – challenges
Last year, I had the opportunity to participate in the first workshop dedicated to forest robotics, held in Montreal or at least that was the intention. At first, there was a request for presentations on that research topic from forest research community, but it turned out that there was not so much feedback on that. Perhaps forest research community is reluctant to modern technology. So instead, the workshop brought together forestry professionals and others like software-, electrical- and mechanical- engineers as well as private companies presenting their successful implementation of robot technology. It turns out that many industries are already using unmanned machines to assist humans and tele-operations are very common practice.
My impression was that there was a miscommunication between those two parties. Moravec paradox that I have cited at the beginning of this text explains the general concept. Forestry professionals request the machines to perform advanced tasks but those simple ones are the most difficult and need more attention.
Perception of the forest is difficult. People studying forest science realize that many of forest properties are not quantifiable and very often rule of thumb is applied when it comes to forest management and planning activities. That is why all actions that we undertake have to undergo more intuitive decision-making.
3D map of forest generated using Simultaneous Localization and Mapping together with trajectory where the robot was driving. Map’s author: Marek Pierzchała
First, forest as environment is unstructured and that makes it difficult to perceive. One of the main tasks in mobile robotics is localization and navigation, which are very complex issues. Localization is a task of finding an object position in the surrounding scene. At this moment Global Navigation Satellite System (GNSS) is not a feasible solution for localization in the forest because of bad signal reception. It is however exploited for adding spatial reference to machine log data but is not enough to precisely locate the machine or system elements. Therefore, there is still a lot to do in that field. Localization without GPS is possible with use of sensors like wheels (that provide odometry) supported by other sensors like Inertial Measurement Unit (IMU) as well as heavily exploited cameras, laser scanners and depth sensors. Combination of different measurements can help in precise localization of the machine together with building a map where free and occupied space is mapped. This process is supported by a dedicated software utilizing robot algorithms for example computer vision algorithms, to detect landmarks, that could be used for reference. In artificial environment like cities, we can observe distinct features that can be used for data association (for example recognizing place), as well as flat surfaces, planes that make a robot’s life easier. Nevertheless, when we put a robot in the forest, he gets lost like a child, because all the trees look the same, and even if not, their appearance is not significant enough to distinguish. Apparently, the problem is deeper.
Another extremely difficult task is a mobility. I mentioned wheel odometry that is a reliable source for localization but not in the forest due to rough terrain and wheel’s slip. But even if we use them, it is still hard to penetrate the forest autonomously.
What would we mount under the chassis? Would it be wheels, continuous track or perhaps legs? Size matters but definitely it is a challenge as drive-ability is very subjective and it depends on operator skills or his courage. Autonomous machine would have difficulties in assessing what is drivable and if the obstacle can be ignored or should be omitted.
There are successful implementations already. Vision based systems for stack measurement, UAV’s acquiring imagery for surface reconstruction and airborne patrol systems, advanced control systems for harvesters etc.
Still there is a lot to do in the field of automation in forest, before we can let the machines to work unassisted.
Check out how robots from Boston dynamics can already walk in the forest:
Atlas, The Next Generation
Author of the post:
Marek Pierzchala – Forest engineer, PhD Candidate, graduated from Cracow Agricultural University and University of Life Sciences in Vienna. Working at Norwegian Institute for Bioeconomy Research. Mainly focuses on use of senors (mainly imaging sensors and laser) and robot mapping.
Winter sports enthusiast.