Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Diffusion Spline-based Navigation Policy in Dynamic Environments
Diffusion models have a huge potential in motion planning and navigation. In this project, we focus on generating spline-based trajectories using diffusion models able to make ANYmal navigate in extremely challenging dynamic environments
Diffusion models have been recently proposed for motion planning and navigation (e.g., [1-5]). In many of these works, however, the trajectory produced by the diffusion process is a sequence of waypoints, and the environment is static. The goal of this project is twofold. On one hand it aims to explore the advantages of using instead the control points of a spline (in terms of computation time, guaranteed smoothness,...). On the other hand, this project plans to study how this spline-based diffusion model can be used to plan multimodal trajectories in dynamic environments for ANYmal [6,7]. Imposition of constraints in the spline will also be considered [8].
----------------------------
[1] Reuss, Moritz, et al. "Goal-conditioned imitation learning using score-based diffusion policies." arXiv preprint arXiv:2304.02532 (2023).
[2] Chi, Cheng, et al. "Diffusion policy: Visuomotor policy learning via action diffusion." arXiv preprint arXiv:2303.04137 (2023).
[3] Janner, Michael, et al. "Planning with diffusion for flexible behavior synthesis." arXiv preprint arXiv:2205.09991 (2022).
[4] Carvalho, J.; Le, A.T.; Baierl, M.; Koert, D.; Peters, J. (2023). Motion Planning Diffusion: Learning and Planning of Robot Motions with Diffusion Models, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).
[5] Carvalho, J.; Baierl, M; Urain, J; Peters, J. (2022). Conditioned Score-Based Models for Learning Collision-Free Trajectory Generation, NeurIPS 2022 Workshop on Score-Based Methods.
[6] Tordesillas, Jesus, and Jonathan P. How. "MADER: Trajectory planner in multiagent and dynamic environments." IEEE Transactions on Robotics 38.1 (2021): 463-476.
[7] Tordesillas, Jesus, and Jonathan P. How. "Deep-PANTHER: Learning-based perception-aware trajectory planner in dynamic environments." IEEE Robotics and Automation Letters 8.3 (2023): 1399-1406.
[8] Tordesillas, Jesus, Jonathan P. How, and Marco Hutter. "RAYEN: Imposition of Hard Convex Constraints on Neural Networks." arXiv preprint arXiv:2307.08336 (2023).
Diffusion models have been recently proposed for motion planning and navigation (e.g., [1-5]). In many of these works, however, the trajectory produced by the diffusion process is a sequence of waypoints, and the environment is static. The goal of this project is twofold. On one hand it aims to explore the advantages of using instead the control points of a spline (in terms of computation time, guaranteed smoothness,...). On the other hand, this project plans to study how this spline-based diffusion model can be used to plan multimodal trajectories in dynamic environments for ANYmal [6,7]. Imposition of constraints in the spline will also be considered [8].
----------------------------
[1] Reuss, Moritz, et al. "Goal-conditioned imitation learning using score-based diffusion policies." arXiv preprint arXiv:2304.02532 (2023).
[2] Chi, Cheng, et al. "Diffusion policy: Visuomotor policy learning via action diffusion." arXiv preprint arXiv:2303.04137 (2023).
[3] Janner, Michael, et al. "Planning with diffusion for flexible behavior synthesis." arXiv preprint arXiv:2205.09991 (2022).
[4] Carvalho, J.; Le, A.T.; Baierl, M.; Koert, D.; Peters, J. (2023). Motion Planning Diffusion: Learning and Planning of Robot Motions with Diffusion Models, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).
[5] Carvalho, J.; Baierl, M; Urain, J; Peters, J. (2022). Conditioned Score-Based Models for Learning Collision-Free Trajectory Generation, NeurIPS 2022 Workshop on Score-Based Methods.
[6] Tordesillas, Jesus, and Jonathan P. How. "MADER: Trajectory planner in multiagent and dynamic environments." IEEE Transactions on Robotics 38.1 (2021): 463-476.
[7] Tordesillas, Jesus, and Jonathan P. How. "Deep-PANTHER: Learning-based perception-aware trajectory planner in dynamic environments." IEEE Robotics and Automation Letters 8.3 (2023): 1399-1406.
[8] Tordesillas, Jesus, Jonathan P. How, and Marco Hutter. "RAYEN: Imposition of Hard Convex Constraints on Neural Networks." arXiv preprint arXiv:2307.08336 (2023).
- Literature review
- Implementation
- Testing on the robot
- Literature review - Implementation - Testing on the robot
- Knowledge of Splines
- Knowledge of diffusion models
- Programming experience in Python
- Knowledge of Splines - Knowledge of diffusion models - Programming experience in Python
Please send your CV, transcripts, and a short motivational statement to Jesus Tordesillas (jtordesillas@ethz.ch) and CC Jonas Frey (jonfrey@ethz.ch). If you have a github/bitbucket/... account, please include a link to it in your CV
Please send your CV, transcripts, and a short motivational statement to Jesus Tordesillas (jtordesillas@ethz.ch) and CC Jonas Frey (jonfrey@ethz.ch). If you have a github/bitbucket/... account, please include a link to it in your CV
