Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Motion-Primitives Planning for Legged Robots
Autonomous navigation in complex environments is a fundamental problem in robotics and typically requires samples and search on continuous space for traversable paths. When navigating through areas without environmental priors, re-sampling or searching exclusively through updated traversable space is expensive with onboard sensing, limiting the execution of dynamic motions. In this project, we aim to explore methods to generate trajectories for legged robots toward highly efficient and reactive local planning in complex environments.
Autonomous navigation in complex environments is a fundamental problem in robotics and typically requires samples and search on continuous space for traversable paths. When navigating through areas without environmental priors, re-sampling or searching exclusively through updated traversable space is expensive with onboard sensing, limiting the execution of dynamic motions. Current motion-primitives methods for local planning [1, 2] show improvement of online efficiency by pre-computing a kinodynamic-feasible trajectory library to be searched on the fly for collisions-free paths. Such methods have been widely deployed on drones and wheeled-vehicles, but they do not apply directly to agile, holonomic legged robots. For instance, the orientation control of the legged robot is decoupled from its linear motion, and then the trajectory library generated only in three-dimensional euclidean space ℝ3 is not sufficient for it to fit through narrow corridors or staircases. In this project, we aim to explore methods to generate trajectories for legged robots towards highly efficient and reactive local planning in complex environments.
[1] J. Zhang et al. Falco: Fast Likelihood-based Collision Avoidance with Extension to Human-guided Navigation. Journal of Field Robotics, 2020.
[2] G. G. Waibel, et al. "How Rough Is the Path? Terrain Traversability Estimation for Local and Global Path Planning," in IEEE Transactions on Intelligent Transportation Systems
Autonomous navigation in complex environments is a fundamental problem in robotics and typically requires samples and search on continuous space for traversable paths. When navigating through areas without environmental priors, re-sampling or searching exclusively through updated traversable space is expensive with onboard sensing, limiting the execution of dynamic motions. Current motion-primitives methods for local planning [1, 2] show improvement of online efficiency by pre-computing a kinodynamic-feasible trajectory library to be searched on the fly for collisions-free paths. Such methods have been widely deployed on drones and wheeled-vehicles, but they do not apply directly to agile, holonomic legged robots. For instance, the orientation control of the legged robot is decoupled from its linear motion, and then the trajectory library generated only in three-dimensional euclidean space ℝ3 is not sufficient for it to fit through narrow corridors or staircases. In this project, we aim to explore methods to generate trajectories for legged robots towards highly efficient and reactive local planning in complex environments.
[1] J. Zhang et al. Falco: Fast Likelihood-based Collision Avoidance with Extension to Human-guided Navigation. Journal of Field Robotics, 2020. [2] G. G. Waibel, et al. "How Rough Is the Path? Terrain Traversability Estimation for Local and Global Path Planning," in IEEE Transactions on Intelligent Transportation Systems
1. Literature review on local and motion-primitives planning algorithms
2. Research and Implement the planning algorithm for legged robots
3. Integration with ANYmal planning and navigation framework for fast local planning
4. Real-world evaluation of implemented algorithms
1. Literature review on local and motion-primitives planning algorithms 2. Research and Implement the planning algorithm for legged robots 3. Integration with ANYmal planning and navigation framework for fast local planning 4. Real-world evaluation of implemented algorithms
1. Highly motivated for the topic
2. Programming experience (C++/Python/Matlab)
3. Knowledge in planning, perception, and robot dynamics is a plus
1. Highly motivated for the topic 2. Programming experience (C++/Python/Matlab) 3. Knowledge in planning, perception, and robot dynamics is a plus
Fan Yang: fanyang1@ethz.ch
Lorenz Wellhausen: lorenwel@ethz.ch
Matias Mattamala: matias@robots.ox.ac.uk
Fan Yang: fanyang1@ethz.ch Lorenz Wellhausen: lorenwel@ethz.ch Matias Mattamala: matias@robots.ox.ac.uk