Ronen Brafman – Ben-Gurion University
When: 08.06.2022 at 15:00
Abstract: While one-trick robots are abundant in industry,
constructing an autonomous robot that can perform multiple
tasks flexibly, such as a service robot, is a difficult task, we are
yet to reach. A large number of robotics researchers have been working hard to design
and implement algorithms for diverse robotic skills. But these skills are
hard to put together: A lot of software engineering and thought goes
into each script that operates them. In this talk I will present a research
program seeking to change this situation, providing
plug’s play autonomy for robotic via what we call the autonomous operating system (AOS).
This system uses a language we designed for documenting robotic skill code to provide
the AOS with the information needed to decide when and how tol execute each skill.
No additional software-engineering effort is required beyond this specification.
I will describe our initial version of this system and our future plans.
The talk is non-technical and requires little background.
You can see the seminar here
When: Wednesday, April 27th at 15:00
Abstract: In this talk, I first discuss how we leverage machine learning methods to generate cooperative policies for multi-robot systems. I describe how we use Graph Neural Networks (GNNs) to learn effective communication strategies for decentralized coordination. I then show how our GNN-based policy is able to achieve near-optimal performance across a variety of problems, at a fraction of the real-time computational cost. Finally, I present some pioneering real-robot experiments that demonstrate the transfer of our methods to the physical world.
When: Wednesday, April 13th at 15:00
Talk Abstract The questions of dexterity, agility, and learning from a few demonstrations have intrigued robotics researchers. In this talk, I will explore answers and solutions to these questions via the following case studies: (i) a dexterous manipulation system capable of re-orienting novel objects. (ii) a quadruped robot that is substantially more agile than its counterparts (runs, spins) on challenging natural terrains. (iii) framework for learning task-sensitive perceptual representations for planning and out-of-distribution generalization. While a lot of recent progress in robotics is driven by perception, we show that learned controllers can help address problems that were previously thought to be hard. I will discuss our findings, the insights we gained, and the road ahead.
When: Wednesday, April 6th at 15:00
Abstract: Classic state estimation tools (e.g., determining position/velocity of a robot from noisy sensor data) have been in use since the 1960s, perhaps the most famous technique being the Kalman filter. For difficult-to-model nonlinear systems with rich sensing (e.g., almost any real-world robot), clever adaptations are needed to the classic tools. In this talk, I will first briefly summarize an idea that has become standard practice in our group over the last several years: continuous-time trajectory estimation (and its connection to sparse Gaussian process regression). I will then discuss two new frameworks we have been pursuing lately: exactly sparse Gaussian variationally inference (ESGVI) and Koopman state estimation (KoopSE). ESGVI seeks to minimize the Kullback-Leibler divergence between a Gaussian state estimate and the full Bayesian posterior; however, the framework also easily allows for parameter learning through Expectation Maximization and we’ve used this to learn simple parameters such as constant system matrices and covariances, but also to model rich sensors using Deep Neural Networks and learn the weights from data. KoopSE takes a different approach by lifting a nonlinear system into a high-dimensional Reproducing Kernel Hilbert Space where we can treat it as linear and apply classic estimation tools; it also allows for the system to be learned from training data quite efficiently. I will give simple intuitive explanations of the mathematics and show some examples of things working in practice.
Work towards PhD degree under the supervision of Prof. Amir Degani
When: 14.4.2022 at 11:00
The seminar will be held in hybrid format: Water institute auditorium
Abstract: In recent years, there has been considerable advancement in the fields of planning and control of legged robots, but legged robots are still extremely limited. Most research effort has been directed toward bipedal walking and quadrupedal walking and running, and not so much at bipedal running. Dynamic-legged locomotion is difficult to plan and execute, it requires robust control strategies and fast planning algorithms. In this work, we describe a strategy for traversing uneven complex terrain without the need to perform explicit foot placement. This will be achieved using a funnel library approach in combination with an online receding horizon planner. Thus, enabling the robot to take full advantage of its dynamics and the potential reaction forces presented by the environment. We demonstrate, in simulation, multiple scenarios, from traversing a low-friction obstacle, all the way to inclined surfaces and chute climbing with intermittent low-friction patches. We present an efficient approach for generating transitions from running to climbing without the need for a long planning horizon. We investigate the SLIP, ASLIP, and the actuated SLIP models and several in-stride control approaches.
When: 23.3.2022 at 15:00
Abstract: In 2016, Schlumberger started to deploy automated drilling, using plan-based control, across multiple rigs in several basins. Five years later, the technology has drilled more than half-a-million feet under fully autonomous control, with an estimated 17% efficiency improvement over human operations. A significant challenge in this deployment has been the range of rigs on which the system has been deployed, each with different degrees of access to control systems, different sensors and all of them in different geophysical settings. In this talk I will outline the planning technology and the benefits it has brought in the achievement of this success, the approaches we have adopted to interfacing between planning and execution, the scope of some of the other uses we have found for the same approach and, depending on time, some of the planning-related problems we are currently investigating.
Work towards MSc degree under the supervision of Dr. Oren Salzman
When: 7.3.2022 at 9:30
Abstract: In Lifelong Multi-Agent Path Finding (L-MAPF) a team of agents performs a stream of tasks consisting of multiple locations to be visited by the agents on a shared graph while avoiding collisions with one another. L-MAPF is typically tackled by partitioning it into multiple consecutive, and hence similar, “one-shot” MAPF queries with a single task assigned to each agent, as in the Rolling-Horizon Collision Resolution (RHCR) algorithm. Therefore, a solution to one query informs the next query, which leads to similarity with respect to the agents’ start and goal positions, and how collisions between the agents need to be resolved from one query to the next. Thus, experience from solving one MAPF query can potentially be used to speedup solving the next one and reduce runtime in L-MAPF overall. Despite this intuition, current L-MAPF planners solve consecutive MAPF queries from scratch. In this paper, we introduce a new RHCR-inspired approach called exRHCR, which exploits experience in its constituent MAPF queries. In particular, exRHCR employs a new extension of Priority-Based Search (PBS), a state-of-the-art MAPF solver.
Our extension, called exPBS, allows to warm-start the search with the priorities between agents used by PBS in the previous MAPF instances. We demonstrate empirically that exRHCR solves L-MAPF up to 25% faster than RHCR, and allows to increase throughput for given task streams by as much as 3%-16% by increasing the number of agents we can cope with for a given time budget.
You can see the seminar here
Work towards MSc degree under the supervision of Prof. Vadim Indelman
When: 24.1.2022 at 14:00
Abstract: One of the most complex tasks of decision making and planning is to gather information. This task becomes even more complex when the state is high-dimensional and its belief cannot be expressed with a parametric distribution. Although the state is high-dimensional, in many problems only a small fraction of it might be involved in transitioning the state and generating observations. We exploit this fact to calculate an information-theoretic expected reward, mutual information (MI), over a much lower-dimensional subset of the state, to improve efficiency and without sacrificing accuracy. A similar approach was used in previous works, yet specifically for Gaussian distributions, and we here extend it for general distributions. Moreover, we apply the dimensionality reduction for cases in which the new states are augmented to the previous, yet again without sacrificing accuracy. We then continue by developing an estimator for the MI which works in a Sequential Monte Carlo (SMC) manner, and avoids the reconstruction of future belief’s surfaces. Finally, we show how this work is applied to the informative planning optimization problem. This work is then evaluated in a simulation of an active SLAM problem, where the improvement in both accuracy and timing is demonstrated.
You can see the seminar here