Search for: All records

Neural programs are highly accurate and structured policies that perform algorith- mic tasks by controlling the behavior of a computation mechanism. Despite the potential to increase the interpretability and the compositionality of the behavior of artificial agents, it remains difficult to learn from demonstrations neural networks that represent computer programs. The main challenges that set algorithmic do- mains apart from other imitation learning domains are the need for high accuracy, the involvement of specific structures of data, and the extremely limited observabil- ity. To address these challenges, we propose to model programs as Parametrized Hierarchical Procedures (PHPs). A PHP is a sequence of conditional operations, using a program counter along with the observation to select between taking an elementary action, invoking another PHP as a sub-procedure, and returning to the caller. We develop an algorithm for training PHPs from a set of supervisor demonstrations, only some of which are annotated with the internal call structure, and apply it to efficient level-wise training of multi-level PHPs. We show in two benchmarks, NanoCraft and long-hand addition, that PHPs can learn neural pro- grams more accurately from smaller amounts of both annotated and unannotated demonstrations. 

An option is a short-term skill consisting of a control policy for a specified region of the state space, and a termination condition recognizing leaving that region. In prior work, we proposed an algorithm called Deep Discovery of Options (DDO) to discover options to accelerate reinforcement learning in Atari games. This paper studies an extension to robot imitation learning, called Discovery of Deep Continuous Options (DDCO), where low-level continuous control skills parametrized by deep neural networks are learned from demonstrations. We extend DDO with: (1) a hybrid categorical–continuous distribution model to parametrize high-level policies that can invoke discrete options as well continuous control actions, and (2) a cross-validation method that relaxes DDO’s requirement that users specify the number of options to be discovered. We evaluate DDCO in simulation of a 3-link robot in the vertical plane pushing a block with friction and gravity, and in two physical experiments on the da Vinci surgical robot, needle insertion where a needle is grasped and inserted into a silicone tissue phantom, and needle bin picking where needles and pins are grasped from a pile and categorized into bins. In the 3-link arm simulation, results suggest that DDCO can take 3x fewer demonstrations to achieve the same reward compared to a baseline imitation learning approach. In the needle insertion task, DDCO was successful 8/10 times compared to the next most accurate imitation learning baseline 6/10. In the surgical bin picking task, the learned policy successfully grasps a single object in 66 out of 99 attempted grasps, and in all but one case successfully recovered from failed grasps by retrying a second time. 

Abstract Ultra-pure NaI(Tl) crystals are the key element for a model-independent verification of the long standing DAMA result and a powerful means to search for the annual modulation signature of dark matter interactions. The SABRE collaboration has been developing cutting-edge techniques for the reduction of intrinsic backgrounds over several years. In this paper we report the first characterization of a 3.4 kg crystal, named NaI-33, performed in an underground passive shielding setup at LNGS. NaI-33 has a record low $$^{39}$$ 39 K contamination of 4.3 ± 0.2 ppb as determined by mass spectrometry. We measured a light yield of 11.1 ± 0.2 photoelectrons/keV and an energy resolution of 13.2% (FWHM/E) at 59.5 keV. We evaluated the activities of $$^{226}$$ 226 Ra and $$^{228}$$ 228 Th inside the crystal to be $$5.9\pm 0.6~\upmu $$ 5.9 ± 0.6 μ Bq/kg and $$1.6\pm 0.3~\upmu $$ 1.6 ± 0.3 μ Bq/kg, respectively, which would indicate a contamination from $$^{238}$$ 238 U and $$^{232}$$ 232 Th at part-per-trillion level. We measured an activity of 0.51 ± 0.02 mBq/kg due to $$^{210}$$ 210 Pb out of equilibrium and a $$\alpha $$ α quenching factor of 0.63 ± 0.01 at 5304 keV. We illustrate the analyses techniques developed to reject electronic noise in the lower part of the energy spectrum. A cut-based strategy and a multivariate approach indicated a rate, attributed to the intrinsic radioactivity of the crystal, of $$\sim $$ ∼ 1 count/day/kg/keV in the [5–20] keV region. 

Abstract SABRE is a dark matter direct detection experiment aiming to measure the annual modulation of the dark matter interaction rate in NaI(Tl) crystals. SABRE focuses on the achievement of an ultra-low background rate operating high-purity NaI(Tl) crystals in a liquid scintillator veto for active background rejection. Moreover, twin experiments will be located in both Northern and Southern hemispheres (Italy and Australia) to disentangle any possible contribution from seasonal or site-related effects. In this article the results of the first measurements with a NaI(Tl) crystal for the SABRE experiment performed at LNGS are presented.