DeepSoMo improves deformable robot motion estimates from internal sensors
DeepSoMo: Deformable Object Motion Estimation under Proprioceptive Partial Observability via Predictive Latent Dynamics
Robotics
Summary
Estimating how soft, bendy robots move using only their internal sensors is very hard because these sensors only see part of what is going on. To guess the robot's full shape and motion over time, you need a method that remembers past movements and predicts future ones. The authors created DeepSoMo, a system that learns to represent the robot's movements in a way that stays consistent and helps predict what's next. Tests using detailed simulations show DeepSoMo is better at tracking the robot’s position, shape, and contact with objects than methods with similar design.
What this means in practice
- •For soft robotics engineers: Estimate the motion and shape of soft robots over time using only internal sensor data for improved control and monitoring.
- •For robotics simulation developers: Use DeepSoMo’s latent dynamic models to enhance simulation realism for deformable robots with partial internal observations.
Authors
Hanwen Zhao, Weicheng Dai, Luoyao Chen, Chen Feng
Abstract
Estimating the motion of deformable bodies in soft robotics from proprioceptive sensing is challenging due to partial observability, as internal observations provide only indirect access to global motion, contact, and shape evolution. In such settings, instantaneous observations are insufficient, and accurate estimation requires temporally consistent latent representations. This paper studies deformable motion estimation under proprioceptive partial observability from a latent dynamics perspective. We argue that a key requirement for long-horizon estimation is latent temporal consistency, where latent states remain predictive of future observations. To this end, we present DeepSoMo, a learning-based framework that enforces predictive structure in latent dynamics through a self-supervised latent prediction constraint. We evaluate the approach in a canonical deformable dynamics setting using high-fidelity finite-element simulation as a controlled source of ground truth. Results demonstrate consistent improvements in trajectory estimation, orientation recovery, shape reconstruction, and contact inference over architecture-matched recurrent baselines.