Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
End-to-end Leaning Terrain Cost with Robot Kinematic Constraints
Navigating the unpredictable off-road environment, autonomous robots require a tailored approach to overcome obstacles and optimize pathfinding. Our proposed terrain cost mapping system goes beyond traditional processing by factoring in each robot's specific kinematic abilities. We introduce a novel simulation-based Roll-Out technique to predict a robot's stability over varied terrains, thereby calculating a precise terrain cost. This innovative strategy promises to enhance autonomous navigation by ensuring safe and efficient traversal tailored to individual robotic capabilities.
Autonomous navigation in off-road terrains poses unique challenges due to the absence of clear pathways and the presence of obstacles like rocks, fallen trees, and uneven ground, which demand advanced perception and decision-making capabilities from robots. Unlike in structured environments, the diverse off-road terrain types significantly influence a robot's mobility, necessitating a navigation strategy tailored to each robot's kinematic abilities, such as wheel size, suspension type, and motor power. Recognizing the varying 'cost' of terrain traversal is crucial for optimizing path planning.
To address this, we propose a nuanced terrain cost mapping system that not only processes LiDAR data but also integrates it with the robot's physical capabilities, ensuring informed path choices. A promising approach we plan to explore involves vehicle Roll-Out in simulation. By simulating the robot's interaction with randomly generated terrain, considering factors like friction, terrain geometry, and the vehicle's kinematic configuration, we can assess stability and determine the associated terrain cost. This method offers a potential breakthrough in accurately predicting the feasibility of navigation maneuvers, thereby ensuring the robot's stability and safety in dynamic off-road conditions.
Autonomous navigation in off-road terrains poses unique challenges due to the absence of clear pathways and the presence of obstacles like rocks, fallen trees, and uneven ground, which demand advanced perception and decision-making capabilities from robots. Unlike in structured environments, the diverse off-road terrain types significantly influence a robot's mobility, necessitating a navigation strategy tailored to each robot's kinematic abilities, such as wheel size, suspension type, and motor power. Recognizing the varying 'cost' of terrain traversal is crucial for optimizing path planning.
To address this, we propose a nuanced terrain cost mapping system that not only processes LiDAR data but also integrates it with the robot's physical capabilities, ensuring informed path choices. A promising approach we plan to explore involves vehicle Roll-Out in simulation. By simulating the robot's interaction with randomly generated terrain, considering factors like friction, terrain geometry, and the vehicle's kinematic configuration, we can assess stability and determine the associated terrain cost. This method offers a potential breakthrough in accurately predicting the feasibility of navigation maneuvers, thereby ensuring the robot's stability and safety in dynamic off-road conditions.
- Kinematic Modelling
- Off-road Simulation
- Terrain Neural Network Design
- Integration and Deployment on Real-Hardware
- Kinematic Modelling - Off-road Simulation - Terrain Neural Network Design - Integration and Deployment on Real-Hardware
- Good Coding / Learning Experience (PyTorch)
- Basic Understanding of Robot Kinematic and Dynamics Modeling
- Motivated by the topic
- Good Coding / Learning Experience (PyTorch) - Basic Understanding of Robot Kinematic and Dynamics Modeling - Motivated by the topic
- Fan Yang (fanyang1@ethz.ch)
- Jonas Frey (jonfrey@ethz.ch)
- Fan Yang (fanyang1@ethz.ch) - Jonas Frey (jonfrey@ethz.ch)
