FORWARD: Dataset of a forwarder operating in rough terrain

Area of Science:

• Forestry Science
• Robotics
• Computer Vision

Background:

• Autonomous operation of forest machinery is crucial for efficiency and safety.
• High-resolution multimodal datasets are needed for developing advanced AI algorithms in this domain.

Purpose of the Study:

• To present the FORWARD dataset, a comprehensive collection of sensor data from a forestry forwarder.
• To facilitate the development of AI models for trafficability, perception, and autonomous control of forest machines.

Main Methods:

• Collected high-resolution multimodal data from a Komatsu forwarder using telematics, cameras, IMUs, and operator sensors.
• Annotated 18 hours of 360° video data with work elements and included StanForD production logs.
• Conducted experiments with varying conditions (steel tracks, load weights, speeds) on forest roads and terrain.

Main Results:

• The FORWARD dataset provides centimeter-accurate positioning, detailed sensor logs, and extensive video material.
• Includes annotated work elements and scenario specifications for diverse research applications.
• Data is made publicly available for long-term use in research and development.

Conclusions:

• The FORWARD dataset is a valuable resource for advancing AI in forestry, particularly for autonomous control, perception, and trafficability.
• Enables research into efficiency, fuel consumption, safety, and environmental impact of forest operations.
• Supports the auto-generation and calibration of forestry machine simulators and automation scenarios.