SymmGrid: Super-Scaling On-Robot Learning with Parallelized Symmetries and Egocentric-Exocentric Visual Perception

2026-07-29

AI Summary

This work addresses the challenge of sample inefficiency in real-world robotic reinforcement learning. The authors propose SymmGrid, a novel framework that, for the first time, integrates parallel symmetric transformations and hyperscaling into physical robot learning. By constructing a symmetric Markov decision process with a geometric grid structure, SymmGrid fuses visual and proprioceptive inputs at the trajectory level and leverages homography transformations to preserve spatial consistency, enabling highly effective data augmentation. Symmetric experiences are generated directly within the replay buffer, substantially improving sample efficiency and generalization. Evaluated on peg insertion, cable routing, and object manipulation tasks, the method achieves 1.37–2.17× faster training convergence and 1.09–1.27× higher success rates, with the fastest task trained in only 7.9 minutes and up to a 2.59× improvement in normalized Area Under the Curve (nAUC).

Abstract

Deep reinforcement policy learning directly in physical robots (on-robot learning) remains bottlenecked by slow wall-clock training times. We present SymmGrid, a trajectory level augmentation framework inspired by parallelized symmetries that super-scales group transformations to significantly accelerate on-robot learning in both egocentric and exocentric visual setups. We model a Markov Decision Process (MDP) under a symmetry tree, in which state-action pairs have admissible parallelized invariant transformations that yield a geometric grid structure. The state is modelled with ego- or exocentric images and proprioception information. The latter require special treatment, in the form of homographies, to warp visual scenes in line with their corresponding spatial transformations. These parallelized transformations produce a large set of unique symmetric equivalences that populate the replay buffer with diverse and consistent experiences that speed up learning and improve performance. We present extensive training and evaluations performed directly on real robot manipulation contact tasks including peg-insertions, cable routing, and object relocations. Relative to SOTA, SymmGrid achieved wall-clock training convergence speed-ups of 1.37-2.17x, evaluation success rate improvements of 1.09x-1.27x, fastest training convergence times of 16.6, 10.9, and 79.3 minutes respectively. For trajectory wide assessments, we used normalized area under the curve (nAUC) ratios. SymmGrid achieved improvements of up to 2.59x. These results confirm that simple branch symmetries can have an outsized result due to super-scaling and bring us closer to sub-10 minute on-robot learning training in manipulation tasks suitable for arms and humanoids. The project page is available at symmgrid-robot.github.io

Problem

• on-robot learning
• wall-clock training time
• reinforcement learning
• robot manipulation
• sample efficiency

Innovation

• parallelized symmetries
• super-scaling
• egocentric-exocentric perception
• trajectory augmentation
• on-robot reinforcement learning