Ergodicity reveals assistance and learning from physical human-robot interaction

Fitzsimons, Kathleen; Acosta, Ana Marı́a; Dewald, Julius P. A.; Murphey, Todd D.

doi:10.1126/scirobotics.aav6079

Cited by 13 publications

(6 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The mid-level control strategies vary significantly across different disciplines since impedance and force control techniques are generally tailored to meet task-specific requirements. These requirements vary from feasible and dynamically equilibrated locomotion controllers (Ugurlu et al, 2016;Ugurlu et al, 2020) to minimization of a user's deficit in human-machine interaction (Fitzsimons et al, 2019). Therefore, synthesis of a mid-level controller requires the determination of task-specific needs in accordance with the target use of the wearable robot.…”

Section: Mid-level Controlmentioning

confidence: 99%

Challenges and solutions for application and wider adoption of wearable robots

et al. 2021

View full text Add to dashboard Cite

The science and technology of wearable robots are steadily advancing, and the use of such robots in our everyday life appears to be within reach. Nevertheless, widespread adoption of wearable robots should not be taken for granted, especially since many recent attempts to bring them to real-life applications resulted in mixed outcomes. The aim of this article is to address the current challenges that are limiting the application and wider adoption of wearable robots that are typically worn over the human body. We categorized the challenges into mechanical layout, actuation, sensing, body interface, control, human–robot interfacing and coadaptation, and benchmarking. For each category, we discuss specific challenges and the rationale for why solving them is important, followed by an overview of relevant recent works. We conclude with an opinion that summarizes possible solutions that could contribute to the wider adoption of wearable robots.

show abstract

Section: Mid-level Controlmentioning

confidence: 99%

Challenges and solutions for application and wider adoption of wearable robots

et al. 2021

View full text Add to dashboard Cite

show abstract

“…4: Comparison of best task executions from 24 novice participants and skill reconstructions based on learned objectives, using only positive demonstrations (posonly, orange) and using both positive and negative demonstrations (posneg, yellow). We employ two performance metrics for the comparison: task success time (left) and the ergodic metric [30] (right), which measures information captured about the task in the learned distributions by comparing it to the true task definition. For both metrics, posonly skill reconstructions achieve performance comparable to or better than the novice demonstrations.…”

Section: B Experimental Platformsmentioning

confidence: 99%

Ergodic imitation: Learning from what to do and what not to do

Kalinowska

Prabhakar

Fitzsimons

et al. 2021

2021 IEEE International Conference on Robotics and Automation (ICRA)

Self Cite

View full text Add to dashboard Cite

With growing access to versatile robotics, it is beneficial for end users to be able to teach robots tasks without needing to code a control policy. One possibility is to teach the robot through successful task executions. However, nearoptimal demonstrations of a task can be difficult to provide and even successful demonstrations can fail to capture task aspects key to robust skill replication. Here, we propose a learning from demonstration (LfD) approach that enables learning of robust task definitions without the need for near-optimal demonstrations. We present a novel algorithmic framework for learning tasks based on the ergodic metric-a measure of information content in motion. Moreover, we make use of negative demonstrations-demonstrations of what not to do-and show that they can help compensate for imperfect demonstrations, reduce the number of demonstrations needed, and highlight crucial task elements improving robot performance. In a proofof-concept example of cart-pole inversion, we show that negative demonstrations alone can be sufficient to successfully learn and recreate a skill. Through a human subject study with 24 participants, we show that consistently more information about a task can be captured from combined positive and negative (posneg) demonstrations than from the same amount of just positive demonstrations. Finally, we demonstrate our learning approach on simulated tasks of target reaching and table cleaning with a 7-DoF Franka arm. Our results point towards a future with robust, data-efficient LfD for novice users.

show abstract

“…Because perfect ergodicity is only possible on infinite time horizons, we require a metric that can be maximized over finite-horizons through decision-making-such a metric was developed in [183]. Metrics on ergodicity provide a principle of motion [13,24] similar to energy minimization and error minimization, and can be used to synthesize automated exploration for learning, as we will see in Section 5. The ergodic metric in [183] provides a method for comparing a trajectory x(t)-a singleton at any given time t-to a distribution Φ(x) through their spatial Fourier transforms.…”

Section: Ergodicitymentioning

confidence: 99%

“…Recent work analyzing animal and human movement has begun to interpret physical bodies as information channels and motions as informationcarrying signals. This has led to the development of methods that help to understand the pathology of conditions such as autism spectrum disorder [21], schizophrenia [22], and stroke [23,24] through an information-theoretic analysis of movement. More generally, this suggests that in order to realize learning objectives, active learning requires measures that capture the information content of an agent's movements.…”

Section: Introductionmentioning

confidence: 99%

Active Learning in Robotics: A Review of Control Principles

Taylor,

Berrueta,

Murphey

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Active learning is a decision-making process. In both abstract and physical settings, active learning demands both analysis and action. This is a review of active learning in robotics, focusing on methods amenable to the demands of embodied learning systems. Robots must be able to learn efficiently and flexibly through continuous online deployment. This poses a distinct set of control-oriented challenges-one must choose suitable measures as objectives, synthesize realtime control, and produce analyses that guarantee performance and safety with limited knowledge of the environment or robot itself. In this work, we survey the fundamental components of robotic active learning systems. We discuss classes of learning tasks that robots typically encounter, measures with which they gauge the information content of observations, and algorithms for generating action plans. Moreover, we provide a variety of examples-from environmental mapping to nonparametric shape estimation-that highlight the qualitative differences between learning tasks, information measures, and control techniques. We conclude with a discussion of control-oriented open challenges, including safety-constrained learning and distributed learning.

show abstract

Ergodicity reveals assistance and learning from physical human-robot interaction

Cited by 13 publications

References 45 publications

Challenges and solutions for application and wider adoption of wearable robots

Challenges and solutions for application and wider adoption of wearable robots

Ergodic imitation: Learning from what to do and what not to do

Active Learning in Robotics: A Review of Control Principles

Contact Info

Product

Resources

About