# Learning the Right Abstraction: Neural Reduced Dynamics for Complex Robot Control

arXiv: https://arxiv.org/abs/2608.19375 · HTML: https://mlsanity.com/paper/2608.19375/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in RL for robot locomotion and manipulation, one of 15 directions in Deep reinforcement learning methods.
- Builds on 22 of its 37 references that are in the map.
- First posted on arXiv 19 Aug 2026.

## Placed in

- Deep reinforcement learning methods / [RL for robot locomotion and manipulation](https://mlsanity.com/deeprl/rl-for-robot-locomotion-and-manipulation/)

## Builds on (most cited first, 22 of 22)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Dream to Control: Learning Behaviors by Latent Imagination](https://mlsanity.com/paper/1912.01603/) (2019)
- [Learning Latent Dynamics for Planning from Pixels](https://mlsanity.com/paper/1811.04551/) (2018)
- [Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models](https://mlsanity.com/paper/1805.12114/) (2018)
- [When to Trust Your Model: Model-Based Policy Optimization](https://mlsanity.com/paper/1906.08253/) (2019)
- [Brax -- A Differentiable Physics Engine for Large Scale Rigid Body Simulation](https://mlsanity.com/paper/2106.13281/) (2021)
- [Linearly-Recurrent Autoencoder Networks for Learning Dynamics](https://mlsanity.com/paper/1712.01378/) (2017)
- [Context-aware Dynamics Model for Generalization in Model-Based Reinforcement Learning](https://mlsanity.com/paper/2005.06800/) (2020)
- [Simplicits: Mesh-Free, Geometry-Agnostic, Elastic Simulation](https://mlsanity.com/paper/2407.09497/) (2024)
- [Learning Terrain-Aware Kinodynamic Model for Autonomous Off-Road Rally Driving With Model Predictive Path Integral Control](https://mlsanity.com/paper/2305.00676/) (2023)
- [Neural Robot Dynamics](https://mlsanity.com/paper/2508.15755/) (2025)
- [Simulation Distillation: Pretraining World Models in Simulation for Rapid Real-World Adaptation](https://mlsanity.com/paper/2603.15759/) (2026)
- [Contact-Aware Neural Dynamics](https://mlsanity.com/paper/2601.12796/) (2026)
- [A Physics-Based Continuum Model for Versatile, Scalable, and Fast Terramechanics Simulation](https://mlsanity.com/paper/2507.05643/) (2025)
- [FreeForm: Reduced-Order Deformable Simulation from Particle-Based Skinning Eigenmodes](https://mlsanity.com/paper/2605.29318/) (2026)
- [FNODE: Flow-Matching for data-driven simulation of constrained multibody systems](https://mlsanity.com/paper/2509.00183/) (2025)
- [Learning Legged MPC with Smooth Neural Surrogates](https://mlsanity.com/paper/2601.12169/) (2026)
- [Nonplanar Model Predictive Control for Autonomous Vehicles with Recursive Sparse Gaussian Process Dynamics](https://mlsanity.com/paper/2602.16206/) (2026)
- [High-Speed, All-Terrain Autonomy: Ensuring Safety at the Limits of Mobility](https://mlsanity.com/paper/2603.20525/) (2026)
- [Adapting Neural Robot Dynamics on the Fly for Predictive Control](https://mlsanity.com/paper/2604.04039/) (2026)
- [Multistep Belief Space Dynamics Learning For Risk-Aware Control](https://mlsanity.com/paper/2605.12628/) (2026)

## Built on by (most cited first, 1 of 1)

- [Verti-WM: A Physics-Aided Exteroceptive World Model for Off-Road Reinforcement Learning](https://mlsanity.com/paper/2609.23118/) (2026)

## Abstract (arXiv)

High-fidelity embodied AI simulators provide realistic evaluation of complex robotic systems, but their computational cost limits their direct use for large-scale reinforcement learning campaigns. We advocate the use of less accurate but more expeditious simulations, which might draw on data-driven, e.g., neural dynamics, models. This contribution argues that the practical value of a neural dynamics model for complex robot control lies in learning the right abstraction: a reduced state that preserves the control-relevant physics of the high-fidelity system while enabling high-throughput policy learning. We develop a neural reduced dynamics (NRD) framework that separates the state the model propagates from what can be supplied as an input or recovered analytically, trains policies entirely inside the frozen learned model, and validates them back in the high-fidelity simulator. Two case studies instantiate it across three control tasks: terrain-aware HMMWV trajectory tracking on rigid, bumpy and deformable Continuum Representation Model (CRM) terrain; and goal reaching for a stock tracked vehicle and its front-mounted articulated arm. Every policy transfers back to the high-fidelity simulator. A single policy trained inside the terrain-conditioned dynamics model, and given no terrain input of its own, attains lower median and mean tracking error than both single-terrain specialists on all three terrains, including zero-shot bumpy terrain. Quantitatively, the tracked vehicle reaches 100 of 100 goals and the arm 97 of 100, with zero contacts or joint-limit violations. The NRD models advance roughly four orders of magnitude faster in simulated time than the high-fidelity simulator scenes they replace, making iterative on-policy learning practical and supporting neural reduced dynamics as a bridge between accurate but expensive physics simulation and scalable robot learning.