Periodic Reporting for period 1 - DIGIFOREST (Digital Analytics and Robotics for Sustainable Forestry)
Reporting period: 2022-09-01 to 2023-08-31
With 43% of the EU covered by forests or woodlands, their well-being is essential for all Europeans. The European forestry industry has been tasked with achieving greater sustainability by the recent EU Forestry Strategy. Targets include improved forest management, sustainable development of wood-based resources and improving the skills of forest workers.
Forests must be managed with a life-cycle view: increasing carbon stocks, protecting biodiversity but also enabling the replacement of carbon-intensive construction materials, such as cement and steel. Current forest management and logging practices fall short of supporting these goals.
In DigiForest, we will propose an innovative novel approach to transform to large-scale precision forestry management. It is centered on the collection of detailed and diverse forest data (at tree level) using mobile robots, its analysis with artificial intelligence, and its presentation to forestry operators to support decision making and robotic selective logging/cutting. Our robot team is diverse: we use both rugged field robots (legged robot ANYmal) as well as more experimental vehicles (various drones to fly also below the canopy). Most ambitious of all is the intention to (semi-)automate a lightweight harvester for sustainable selective logging.
With an improvement in spatial data acquisition, organization and analysis, forestry operators and enterprises can obtain up-to-date, tangible information about the status of their forests down to the individual tree. This will improve their oversight by allowing more accurate growth modelling of forest stands and precise predictions of timber yields. It will remove the uncertainty of when thinning operations are needed or where there are trees which are ready for harvest. It could also enable operators to automatically plan where their staff or equipment should be deployed.
Forests must be managed with a life-cycle view: increasing carbon stocks, protecting biodiversity but also enabling the replacement of carbon-intensive construction materials, such as cement and steel. Current forest management and logging practices fall short of supporting these goals.
In DigiForest, we will propose an innovative novel approach to transform to large-scale precision forestry management. It is centered on the collection of detailed and diverse forest data (at tree level) using mobile robots, its analysis with artificial intelligence, and its presentation to forestry operators to support decision making and robotic selective logging/cutting. Our robot team is diverse: we use both rugged field robots (legged robot ANYmal) as well as more experimental vehicles (various drones to fly also below the canopy). Most ambitious of all is the intention to (semi-)automate a lightweight harvester for sustainable selective logging.
With an improvement in spatial data acquisition, organization and analysis, forestry operators and enterprises can obtain up-to-date, tangible information about the status of their forests down to the individual tree. This will improve their oversight by allowing more accurate growth modelling of forest stands and precise predictions of timber yields. It will remove the uncertainty of when thinning operations are needed or where there are trees which are ready for harvest. It could also enable operators to automatically plan where their staff or equipment should be deployed.
Work has been performed in several directions to ensure the development of the different technologies relevant to the DigiForest objectives for the first year.
As a first step, the members of the consortium have worked in enabling field autonomy of the different robot platforms which will be leveraged in the project. During the first year, the semi-autonomous harvester has received several hardware upgrades to perform chassis balancing and to be able to perform position control of the hydraulic manipulator.
For the ANYmal and the different drones, state-estimation frameworks have been developed to allow for precise odometry in challenging, unstructured environments such as a forest. The state-estimation frameworks rely on LiDAR-IMU sensors or visual-inertial sensors. Also, approaches leveraging submaps for scalable mapping have been developed.
To ensure safe navigation, traversability-aware navigation methods have been developed based either on geometric information or on semantics and geometric information for the different ground-robots. Path planning, based either on submaps or monolithic maps, has been developed for the drones. Also, reactive planning is being studied, to avoid trees and branches of the forest while flying through it, which will allow for autonomous navigation of the drones without the requirement of a prior map.
The BLK laser scanner family from Leica as part of their drone (BLK2FLY) and as handheld carriers (BLK2GO) has also been integrated on the ANYmal. These will be used for high precision measurements of the forests.
A map server interface has been defined to allow for a common representation of the maps generated by the different platforms. These maps will be used to generate the valuable properties related to the forest, after being mapped by the robotic platforms. During this period, work on semantic and object-level mapping in an offline fashion has been done, by post-processing the map information from the map server. Also, geometric modelling is performed on these offline maps to compute estimate the tree trunk via cylindrical primitives.
Two filed trials have been conducted during this period, one in Switzerland and another, more extensive one, in Finland. In these, the different members of the consortium have tested their developments in a real-world scenario and also collected data to continue with their development. With the data collected in these trials, initial work on the development of an individual tree inventory has started and initial efforts on the decision support system development have been performed. Also, a logging planner and interface to allow for the harvester to perform path planning has been developed.
As a first step, the members of the consortium have worked in enabling field autonomy of the different robot platforms which will be leveraged in the project. During the first year, the semi-autonomous harvester has received several hardware upgrades to perform chassis balancing and to be able to perform position control of the hydraulic manipulator.
For the ANYmal and the different drones, state-estimation frameworks have been developed to allow for precise odometry in challenging, unstructured environments such as a forest. The state-estimation frameworks rely on LiDAR-IMU sensors or visual-inertial sensors. Also, approaches leveraging submaps for scalable mapping have been developed.
To ensure safe navigation, traversability-aware navigation methods have been developed based either on geometric information or on semantics and geometric information for the different ground-robots. Path planning, based either on submaps or monolithic maps, has been developed for the drones. Also, reactive planning is being studied, to avoid trees and branches of the forest while flying through it, which will allow for autonomous navigation of the drones without the requirement of a prior map.
The BLK laser scanner family from Leica as part of their drone (BLK2FLY) and as handheld carriers (BLK2GO) has also been integrated on the ANYmal. These will be used for high precision measurements of the forests.
A map server interface has been defined to allow for a common representation of the maps generated by the different platforms. These maps will be used to generate the valuable properties related to the forest, after being mapped by the robotic platforms. During this period, work on semantic and object-level mapping in an offline fashion has been done, by post-processing the map information from the map server. Also, geometric modelling is performed on these offline maps to compute estimate the tree trunk via cylindrical primitives.
Two filed trials have been conducted during this period, one in Switzerland and another, more extensive one, in Finland. In these, the different members of the consortium have tested their developments in a real-world scenario and also collected data to continue with their development. With the data collected in these trials, initial work on the development of an individual tree inventory has started and initial efforts on the decision support system development have been performed. Also, a logging planner and interface to allow for the harvester to perform path planning has been developed.
This has been the first year of the DigiForest project. As for the upcoming years, the objective is to improve the robustness of the different methods which have been developed. This includes: improving the chassis and manipulator control of the harvester; improve the state-etimation methods used in the different platforms; improve the traversability awareness; allow for faster flights in the forest; improving the map instance segmentation and the geometric parametrisation of the tree trunk by fitting more complex shapes.
The decision support system will be further developed. This system will provide non-roboticist users all the necessary information to perform data-driven decisions in the forest; saving time for the forest managers, allowing to reduce costs in their operations, and allowing for more sustainable logging and thinning practices.
In the upcoming years, more field trials will be performed to demonstrate the advanced capabilities of the developed technologies. Through these demonstrations, the consortium will be able to introduce the technology to end-users. It has been identified that this industry is very traditional and reluctant to changes. Therefore, the consortium will have to demonstrate a certain level of maturity and market-readiness. Once this has been achieved, more validations with potential external stakeholders, from both EU and third countries, will be carried out.
The decision support system will be further developed. This system will provide non-roboticist users all the necessary information to perform data-driven decisions in the forest; saving time for the forest managers, allowing to reduce costs in their operations, and allowing for more sustainable logging and thinning practices.
In the upcoming years, more field trials will be performed to demonstrate the advanced capabilities of the developed technologies. Through these demonstrations, the consortium will be able to introduce the technology to end-users. It has been identified that this industry is very traditional and reluctant to changes. Therefore, the consortium will have to demonstrate a certain level of maturity and market-readiness. Once this has been achieved, more validations with potential external stakeholders, from both EU and third countries, will be carried out.