Differentiable Simulation of Hard Contacts with Soft Gradients for Learning and Control
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910064027107328 |
|---|---|
| author | Paulus, Anselm Geist, A. René Schumacher, Pierre Musil, Vít Rappenecker, Simon Martius, Georg |
| author_facet | Paulus, Anselm Geist, A. René Schumacher, Pierre Musil, Vít Rappenecker, Simon Martius, Georg |
| contents | Contact forces introduce discontinuities into robot dynamics that severely limit the use of simulators for gradient-based optimization. Penalty-based simulators such as MuJoCo, soften contact resolution to enable gradient computation. However, realistically simulating hard contacts requires stiff solver settings, which leads to incorrect simulator gradients when using automatic differentiation. Contrarily, using non-stiff settings strongly increases the sim-to-real gap. We analyze penalty-based simulators to pinpoint why gradients degrade under hard contacts. Building on these insights, we propose DiffMJX, which couples adaptive time integration with penalty-based simulation to substantially improve gradient accuracy. A second challenge is that contact gradients vanish when bodies separate. To address this, we introduce contacts from distance (CFD) which combines penalty-based simulation with straight-through estimation. By applying CFD exclusively in the backward pass, we obtain informative pre-contact gradients while retaining physical realism. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_14186 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Differentiable Simulation of Hard Contacts with Soft Gradients for Learning and Control Paulus, Anselm Geist, A. René Schumacher, Pierre Musil, Vít Rappenecker, Simon Martius, Georg Robotics Machine Learning Systems and Control I.2.9; I.2.6; I.6.4; G.1.6 Contact forces introduce discontinuities into robot dynamics that severely limit the use of simulators for gradient-based optimization. Penalty-based simulators such as MuJoCo, soften contact resolution to enable gradient computation. However, realistically simulating hard contacts requires stiff solver settings, which leads to incorrect simulator gradients when using automatic differentiation. Contrarily, using non-stiff settings strongly increases the sim-to-real gap. We analyze penalty-based simulators to pinpoint why gradients degrade under hard contacts. Building on these insights, we propose DiffMJX, which couples adaptive time integration with penalty-based simulation to substantially improve gradient accuracy. A second challenge is that contact gradients vanish when bodies separate. To address this, we introduce contacts from distance (CFD) which combines penalty-based simulation with straight-through estimation. By applying CFD exclusively in the backward pass, we obtain informative pre-contact gradients while retaining physical realism. |
| title | Differentiable Simulation of Hard Contacts with Soft Gradients for Learning and Control |
| topic | Robotics Machine Learning Systems and Control I.2.9; I.2.6; I.6.4; G.1.6 |
| url | https://arxiv.org/abs/2506.14186 |