Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Efficient Simulation of flexible Shanks for the ANYmal Robot
Flexible Materials have the Potential to highly increase the performance of walking robots, such as ANYmal. Unfortunately, using Reinforcement Learning on such Systems, an efficient and stable Simulation Environment for Robots is needed, which is currently not available. The goal is the implementation of a spatial model of a quadruped with flexible shanks as well as writing a simulation framework based on an already existing multi-body simulation for flexible bodies, which includes soft as well as rigid links. All the software will be written in Python. If the project is carried out successfully, the integration of the software stack into NVIDIA's Isaac Gym is possible in cooperation with NVIDIA.
Quadrupedal robots have become extremely
reliable robots with robust control algorithms.
The newest generation of robots is able to
walk fast, to climb stairs or to operate in
difficult terrain. One cornerstone in modern learning-to-walk strategies is an efficient
multi-body dynamics simulation environment
that can cope with frictional contacts. However, most off-the-shelf algorithms are restricted to multi-body systems that are composed of rigid bodies only, thus limiting the robot design to use relatively stiff components.
During the last years, various nonlinear spatial beam formulations and non-smooth numerical time integration schemes have been introduced at the Institute for Nonlinear Mechanics in Stuttgart. The goal of this project is to establish a simulation environment for quadrupeds that includes as additional bodies nonlinear spatial beams. The project is in cooperation with the Institute for Nonlinear Mechanics at the University of Stuttgart (Germany).
Quadrupedal robots have become extremely reliable robots with robust control algorithms. The newest generation of robots is able to walk fast, to climb stairs or to operate in difficult terrain. One cornerstone in modern learning-to-walk strategies is an efficient multi-body dynamics simulation environment that can cope with frictional contacts. However, most off-the-shelf algorithms are restricted to multi-body systems that are composed of rigid bodies only, thus limiting the robot design to use relatively stiff components. During the last years, various nonlinear spatial beam formulations and non-smooth numerical time integration schemes have been introduced at the Institute for Nonlinear Mechanics in Stuttgart. The goal of this project is to establish a simulation environment for quadrupeds that includes as additional bodies nonlinear spatial beams. The project is in cooperation with the Institute for Nonlinear Mechanics at the University of Stuttgart (Germany).
- Introduce benchmark examples to vali-
date intermediate steps
- Evaluate and implement various time-
integration schemes
- Test different beam formulations
- Implement frictional beam-to-ground
contact interaction
- Write a fitting algorithm for precurved
beams
- Introduce benchmark examples to vali- date intermediate steps - Evaluate and implement various time- integration schemes - Test different beam formulations - Implement frictional beam-to-ground contact interaction - Write a fitting algorithm for precurved beams
- Prior knowledge in Computational Dynamics for Robotics and Simulation
- Strong coding skills (Required: Python)
- Strong communication skills and motivation
- Prior knowledge in the simulation of flexible bodies (optionally)
- Prior knowledge in Computational Dynamics for Robotics and Simulation - Strong coding skills (Required: Python) - Strong communication skills and motivation - Prior knowledge in the simulation of flexible bodies (optionally)
Please send an E-Mail with your CV and transcript of records. Within the E-Mail please mention the related projects you have previously worked on.
E-Mail fbjelonic@ethz.ch and cc eugster@inm.uni-stuttgart.de, jonas.harsch@inm.uni-stuttgart.de
Please send an E-Mail with your CV and transcript of records. Within the E-Mail please mention the related projects you have previously worked on. E-Mail fbjelonic@ethz.ch and cc eugster@inm.uni-stuttgart.de, jonas.harsch@inm.uni-stuttgart.de