arXiv subjects

Oleg Sushkov

Publications and source records attributed to Oleg Sushkov.

At least 19 recordsLinked to original sources

arXiv2024

Tunable viscous layers in Corbino geometry using density junctions

In sufficiently clean materials where electron-electron interactions are strong compared to momentum-relaxing scattering processes, electron transport resembles the flow of a viscous fluid. We study hydrodynamic electron transport across density interfaces (n-n junctions) in a 2DEG in the Corbino geometry. From numerical simulations in COMSOL using realistic parameters, we show that we can produce tunable viscous layers at the density interface by varying the density ratio of charge carriers. We quantitatively explain this observation with simple analytic expressions together with boundary conditions at the interface. We also show signatures of these viscous layers in the magnetoresistance. Breaking down viscous and ohmic contributions, we find that when outer radial region of the Corbino has higher charge density compared to the inner region, the viscous layers at the interface serve to suppress the magneto-resistance produced by momentum-relaxing scattering. Conversely, the magneto-resistance is enhanced when the inner region has higher density than the outer. Our results add to the repertoire of techniques for engineering viscous electron flows, which hold a promise for applications in future electronic devices.

cond-mat.mes-hall↗

arXiv2023

RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation

The ability to leverage heterogeneous robotic experience from different robots and tasks to quickly master novel skills and embodiments has the potential to transform robot learning. Inspired by recent advances in foundation models for vision and language, we propose a multi-embodiment, multi-task generalist agent for robotic manipulation. This agent, named RoboCat, is a visual goal-conditioned decision transformer capable of consuming action-labelled visual experience. This data spans a large repertoire of motor control skills from simulated and real robotic arms with varying sets of observations and actions. With RoboCat, we demonstrate the ability to generalise to new tasks and robots, both zero-shot as well as through adaptation using only 100-1000 examples for the target task. We also show how a trained model itself can be used to generate data for subsequent training iterations, thus providing a basic building block for an autonomous improvement loop. We investigate the agent's capabilities, with large-scale evaluations both in simulation and on three different real robot embodiments. We find that as we grow and diversify its training data, RoboCat not only shows signs of cross-task transfer, but also becomes more efficient at adapting to new tasks.

cs.RO↗

arXiv2023

Magnetic response of a two-dimensional viscous electron fluid

It has been established that the Coulomb interactions can transform the electron gas into a viscous fluid. This fluid is realized in a number of platforms, including graphene and two-dimensional semiconductor heterostructures. The defining characteristic of the electron fluid is the formation of layers of charge carriers that are in local thermodynamic equilibrium, as in classical fluids. In the presence of nonuniformities, whirlpools and nontrivial flow profiles are formed, which have been directly imaged in recent experiments. In this paper, we theoretically study the response of the electron fluid to localized magnetic fields. We find that the electric current is suppressed by viscous vortices in regions where magnetic field is sharply varying, causing strong transport signatures. Experimentally, our considerations are relevant since local magnetic fields can be applied to the system through implanting adatoms or embedding micromagnets in the top-gate. Our theory is essential for the characterization and future applications of electron fluids in hydrodynamic spin transport.

cond-mat.mes-hall↗

arXiv2022

Observation of oscillating $g$-factor anisotropy arising from strong crystal lattice anisotropy in GaAs spin-3/2 hole quantum point contacts

Many modern spin-based devices rely on the spin-orbit interaction, which is highly sensitive to the host semiconductor heterostructure and varies substantially depending on crystal direction, crystal asymmetry (Dresselhaus), and quantum confinement asymmetry (Rashba). One-dimensional quantum point contacts are a powerful tool to probe both energy and directional dependence of spin-orbit interaction through the effect on the hole $g$-factor. In this work we investigate the role of cubic crystal asymmetry in driving an oscillation in the in-plane hole $g$-factor anisotropy when the quantum point contact is rotated with respect to the crystal axes, and we are able to separate contributions to the Zeeman Hamiltonian arising from Rashba and cubic crystal asymmetry spin-orbit interactions. The in-plane $g$-factor is found to be extremely sensitive to the orientation of the quantum point contact, changing by a factor of $5$ when rotated by $45^{\circ}$. This exceptionally strong crystal lattice anisotropy of the in-plane Zeeman splitting cannot be explained within axially symmetric theoretical models. Theoretical modelling based on the combined Luttinger, Rashba and Dresselhaus Hamiltonians that we use here reveals new spin-orbit contributions to the in-plane hole $g$-factor and provides an excellent agreement with our experimental data.

cond-mat.mes-hall↗

arXiv2022

Offline Meta-Reinforcement Learning for Industrial Insertion

Reinforcement learning (RL) can in principle let robots automatically adapt to new tasks, but current RL methods require a large number of trials to accomplish this. In this paper, we tackle rapid adaptation to new tasks through the framework of meta-learning, which utilizes past tasks to learn to adapt with a specific focus on industrial insertion tasks. Fast adaptation is crucial because prohibitively large number of on-robot trials will potentially damage hardware pieces. Additionally, effective adaptation is also feasible in that experience among different insertion applications can be largely leveraged by each other. In this setting, we address two specific challenges when applying meta-learning. First, conventional meta-RL algorithms require lengthy online meta-training. We show that this can be replaced with appropriately chosen offline data, resulting in an offline meta-RL method that only requires demonstrations and trials from each of the prior tasks, without the need to run costly meta-RL procedures online. Second, meta-RL methods can fail to generalize to new tasks that are too different from those seen at meta-training time, which poses a particular challenge in industrial applications, where high success rates are critical. We address this by combining contextual meta-learning with direct online finetuning: if the new task is similar to those seen in the prior data, then the contextual meta-learner adapts immediately, and if it is too different, it gradually adapts through finetuning. We show that our approach is able to quickly adapt to a variety of different insertion tasks, with a success rate of 100% using only a fraction of the samples needed for learning the tasks from scratch. Experiment videos and details are available at https://sites.google.com/view/offline-metarl-insertion.

cs.RO↗

arXiv2022

Wish you were here: Hindsight Goal Selection for long-horizon dexterous manipulation

Complex sequential tasks in continuous-control settings often require agents to successfully traverse a set of "narrow passages" in their state space. Solving such tasks with a sparse reward in a sample-efficient manner poses a challenge to modern reinforcement learning (RL) due to the associated long-horizon nature of the problem and the lack of sufficient positive signal during learning. Various tools have been applied to address this challenge. When available, large sets of demonstrations can guide agent exploration. Hindsight relabelling on the other hand does not require additional sources of information. However, existing strategies explore based on task-agnostic goal distributions, which can render the solution of long-horizon tasks impractical. In this work, we extend hindsight relabelling mechanisms to guide exploration along task-specific distributions implied by a small set of successful demonstrations. We evaluate the approach on four complex, single and dual arm, robotics manipulation tasks against strong suitable baselines. The method requires far fewer demonstrations to solve all tasks and achieves a significantly higher overall performance as task complexity increases. Finally, we investigate the robustness of the proposed solution with respect to the quality of input representations and the number of demonstrations.

cs.RO↗

arXiv2021

Robust Multi-Modal Policies for Industrial Assembly via Reinforcement Learning and Demonstrations: A Large-Scale Study

Over the past several years there has been a considerable research investment into learning-based approaches to industrial assembly, but despite significant progress these techniques have yet to be adopted by industry. We argue that it is the prohibitively large design space for Deep Reinforcement Learning (DRL), rather than algorithmic limitations per se, that are truly responsible for this lack of adoption. Pushing these techniques into the industrial mainstream requires an industry-oriented paradigm which differs significantly from the academic mindset. In this paper we define criteria for industry-oriented DRL, and perform a thorough comparison according to these criteria of one family of learning approaches, DRL from demonstration, against a professional industrial integrator on the recently established NIST assembly benchmark. We explain the design choices, representing several years of investigation, which enabled our DRL system to consistently outperform the integrator baseline in terms of both speed and reliability. Finally, we conclude with a competition between our DRL system and a human on a challenge task of insertion into a randomly moving target. This study suggests that DRL is capable of outperforming not only established engineered approaches, but the human motor system as well, and that there remains significant room for improvement. Videos can be found on our project website: https://sites.google.com/view/shield-nist.

cs.AI↗

arXiv2020

S3K: Self-Supervised Semantic Keypoints for Robotic Manipulation via Multi-View Consistency

A robot's ability to act is fundamentally constrained by what it can perceive. Many existing approaches to visual representation learning utilize general-purpose training criteria, e.g. image reconstruction, smoothness in latent space, or usefulness for control, or else make use of large datasets annotated with specific features (bounding boxes, segmentations, etc.). However, both approaches often struggle to capture the fine-detail required for precision tasks on specific objects, e.g. grasping and mating a plug and socket. We argue that these difficulties arise from a lack of geometric structure in these models. In this work we advocate semantic 3D keypoints as a visual representation, and present a semi-supervised training objective that can allow instance or category-level keypoints to be trained to 1-5 millimeter-accuracy with minimal supervision. Furthermore, unlike local texture-based approaches, our model integrates contextual information from a large area and is therefore robust to occlusion, noise, and lack of discernible texture. We demonstrate that this ability to locate semantic keypoints enables high level scripting of human understandable behaviours. Finally we show that these keypoints provide a good way to define reward functions for reinforcement learning and are a good representation for training agents.

cs.RO↗

arXiv2020

Scaling data-driven robotics with reward sketching and batch reinforcement learning

We present a framework for data-driven robotics that makes use of a large dataset of recorded robot experience and scales to several tasks using learned reward functions. We show how to apply this framework to accomplish three different object manipulation tasks on a real robot platform. Given demonstrations of a task together with task-agnostic recorded experience, we use a special form of human annotation as supervision to learn a reward function, which enables us to deal with real-world tasks where the reward signal cannot be acquired directly. Learned rewards are used in combination with a large dataset of experience from different tasks to learn a robot policy offline using batch RL. We show that using our approach it is possible to train agents to perform a variety of challenging manipulation tasks including stacking rigid objects and handling cloth.

cs.RO↗

arXiv2019

Measuring hole g-factor anisotropies using transverse magnetic focusing

Recent theoretical and experimental results from quasi-one dimensional heavy hole systems have suggested that heavy hole gases have a strongly anisotropic g factor. In this theoretical paper, we propose a method for measuring this anisotropy using transverse magnetic focusing (TMF). We demonstrate that for experimentally accessible fields, the g factor anisotropy leads to a relative variation in the characteristic of spin-splitting of the TMF spectrum which allows for the measurement of the anisotropy of the g factor. We show that this variation is insensitive to additional spin-orbit interactions, and is resolvable with current devices.

cond-mat.mes-hall↗

arXiv2018

A Practical Approach to Insertion with Variable Socket Position Using Deep Reinforcement Learning

Insertion is a challenging haptic and visual control problem with significant practical value for manufacturing. Existing approaches in the model-based robotics community can be highly effective when task geometry is known, but are complex and cumbersome to implement, and must be tailored to each individual problem by a qualified engineer. Within the learning community there is a long history of insertion research, but existing approaches are typically either too sample-inefficient to run on real robots, or assume access to high-level object features, e.g. socket pose. In this paper we show that relatively minor modifications to an off-the-shelf Deep-RL algorithm (DDPG), combined with a small number of human demonstrations, allows the robot to quickly learn to solve these tasks efficiently and robustly. Our approach requires no modeling or simulation, no parameterized search or alignment behaviors, no vision system aside from raw images, and no reward shaping. We evaluate our approach on a narrow-clearance peg-insertion task and a deformable clip-insertion task, both of which include variability in the socket position. Our results show that these tasks can be solved reliably on the real robot in less than 10 minutes of interaction time, and that the resulting policies are robust to variance in the socket position and orientation.

cs.RO↗

arXiv2018

Electrical control of the Zeeman spin splitting in two-dimensional hole systems

Semiconductor holes with strong spin-orbit coupling allow all-electrical spin control, with broad applications ranging from spintronics to quantum computation. Using a two-dimensional hole system in a GaAs quantum well, we demonstrate a new mechanism of electrically controlling the Zeeman splitting, which is achieved through altering the hole wave vector $k$. We find a threefold enhancement of the in-plane $g-$factor $g_{\parallel}(k)$. We introduce a new method for quantifying the Zeeman splitting from magnetoresistance measurements, since the conventional tilted field approach fails for two-dimensional systems with strong spin-orbit coupling. Finally, we show that the Rashba spin-orbit interaction suppresses the in-plane Zeeman interaction at low magnetic fields. The ability to control the Zeeman splitting with electric fields opens up new possibilities for future quantum spin-based devices, manipulating non-Abelian geometric phases, and realising Majorana systems in $p-$type superconductor systems.

cond-mat.mes-hall↗

arXiv2017

Interference effects and Huygens' principle in transverse magnetic focusing of electrons and holes

Interference effects form a fundamental pillar of quantum mechanics. In this paper, we examine the interference in spin-orbit coupled transverse magnetic focusing, where a weak magnetic field is used to focus charge carries over mesoscopic scales. We determine a semi-classical form for the Green's function in a weak magnetic field, for the case of both spin-less and spin-orbit coupled charge carriers. The obtained forms for the Greens' function are independent of particle dispersion and are thus applicable to a wide variety of systems.

cond-mat.mes-hall↗

arXiv2015

Magnetic Focussing of Electrons and Holes in the presence of spin orbit interactions

In this work we theoretically investigate transverse magnetic focussing in two dimensional electron and hole gasses with strong spin orbit interactions. We present a general result for spin orbit interaction with singular winding numbers in the adiabatic limit. We then present results for systems with two spin orbit interactions of different winding number appear, using the concrete and experimentally relevant case of an applied in-plane magnetic field in hole systems with Rashba type interactions. We predict that the application of a large in-plane field is found to have a strong effect on the magnetic focussing spectrum.

cond-mat.mes-hall↗

arXiv2015

Asymptotic Freedom in Quantum Antiferromagnet TlCuCl3

Phase transitions in isotropic quantum antiferromagnets are described by an O(3) nonlinear quantum field theory. In three dimensions, the fundamental property of this theory is logarithmic scaling of the coupling constant. At the quantum critical point the coupling asymptotically vanishes and the quasiparticles become free. This logarithmic decay of the coupling constant has never been ob- served. In this paper, we derive finite temperature properties of the field theory and use our results to analyze the existing data on the real antiferromagnet TlCuCl3. Including finite temperatures into the theory, we find that agreement between theory and experiment is sufficiently sensitive to unambiguously identify the asymptotic decay of the coupling constant. We also comment on the unique possibility to study Landau pole-physics in quantum magnets.

cond-mat.str-el↗

arXiv2005

Electron correlations in two-dimensional small quantum dots

We consider circular and elliptic quantum dots with parabolic external confinement, containing 0 - 22 electrons and with values of r_s in the range 0 < r_s < 3. We perform restricted and unrestricted Hartree-Fock calculations, and further take into account electron correlations using second-order perturbation theory. We demonstrate that in many cases correlations qualitatively change the spin structure of the ground state from that obtained under Hartree-Fock and spin-density-functional calculations. In some cases the correlation effects destroy Hund's rule. We also demonstrate that the correlations destroy static spin-density waves observed in Hartree-Fock and spin-density-functional calculations.

cond-mat.mes-hall↗

arXiv1999

Spontaneous dimer order, excitation spectrum, and quantum phase transitions in the J1-J2 Heisenberg model

We overview some recent work and present new results on the ground state properties and the spectrum of excitations of the two-dimensional frustrated Heisenberg antiferromagnet. Spontaneous dimer order is present in the quantum disordered phase of this model. We study the stability and analyze the structure of the spectrum, including the two-particle singlet excitation branch throughout the disordered phase, as well as in the vicinity of the Neel critical point. The variation of the dimer order parameter is also given, and it is argued that near the critical point it reflects the presence of the low-energy singlet bound state.

cond-mat.str-el↗

arXiv1999

Superconducting pairing, and the collective magnetic excitation in the extended 2-dimensional t-J model

To investigate the non-Fermi liquid behavior we consider the extended two-dimensional t-J model which includes additional hopping t''. In the regime t,J << t'' we were able to solve the model analytically. It has a very rich phase diagram including antiferromagnetic (AF) insulator and AF strange metal with different kinds of pseudospin-singlet superconducting pairings (p,d,g-waves). We also demonstrate a collective triplet excitation with energy below the superconducting gap

cond-mat.str-el↗