Search for: All records

Incorporating inductive biases is a promising approach for tackling challenging robot learning domains with sample-efficient solutions. This paper identifies partially observable domains where symmetries can be a useful inductive bias for efficient learning. Specifically, by encoding the equivariance regarding specific group symmetries into the neural networks, our actor-critic reinforcement learning agents can reuse solutions in the past for related scenarios. Consequently, our equivariant agents outperform non-equivariant approaches significantly in terms of sample efficiency and final performance, demonstrated through experiments on a range of robotic tasks in simulation and real hardware. 

Abstract Arc magmas are produced from the mantle wedge, with possible addition of fluids and melts derived from serpentinites and sediments in the subducting slab. Identification of various sources and their relevant contributions to such magmas is challenging; in particular, at continental arcs where crustal assimilation may overprint initial geochemical signatures. This study presents oxygen isotopic compositions of zoned olivine grains from post-caldera basalts and boron contents and isotopes of these basalts and glassy melt inclusions hosted in quartz and clinopyroxene of silicic tuffs in the Toba volcanic system, Indonesia. High-magnesian (≥87 mol% Fo [forsterite]) cores of olivine in the basalts have δ18O values ranging from 5.12‰ to 6.14‰, indicating that the mantle source underneath Toba is variably enriched in 18O. Olivine with <87 mol% Fo has highly variable (4.8–7.2‰), but overall increased, δ18O values, interpreted to reflect assimilation of high δ18O crustal materials during fractional crystallization. Mass balance calculations constrain the overall volume of crustal assimilation for the basalts as ≤13%. The processes responsible for the 18O-enriched basaltic melts are further constrained by boron data that indicate the addition of <0.1 wt% fluids to the mantle, >40% of the fluids being derived from serpentinites and others from altered oceanic crust and sediments. This amount of fluids can increase δ18O of the magma by only ~0.02‰. Approximately 6–9% sediment-derived melt hybridization in the mantle wedge is further needed to yield basaltic melts with δ18O values in equilibrium with those of the high-Fo olivine cores. The cogenetic silicic tuffs, on the other hand, seem to record a higher proportion of fluid addition dominated by sediment-derived fluids to the mantle source, in addition to crustal assimilation. Our reconnaissance study therefore demonstrates the application of combined B and O isotopes to differentiate between melts and fluids derived from serpentinites and sediments in the subducted slab—an application that can be applied to arc magmas worldwide. 

Abstract Real-world grasp detection is challenging due to the stochasticity in grasp dynamics and the noise in hardware. Ideally, the system would adapt to the real world by training directly on physical systems. However, this is generally difficult due to the large amount of training data required by most grasp learning models. In this paper, we note that the planar grasp function is $$\textrm{SE}(2)$$ -equivariant and demonstrate that this structure can be used to constrain the neural network used during learning. This creates an inductive bias that can significantly improve the sample efficiency of grasp learning and enable end-to-end training from scratch on a physical robot with as few as 600 grasp attempts. We call this method Symmetric Grasp learning (SymGrasp) and show that it can learn to grasp “from scratch” in less that 1.5 h of physical robot time. This paper represents an expanded and revised version of the conference paper Zhu et al. (2022).