Towards Collaborative Partners: Design, Shared Control, and Robot Learning for Physical Human-Robot Interaction

ICRA 2024 Workshop

13 May 2024, 9:00 AM - 5:00 PM

Conference Center, Room 411-412

Online Participation: https://us06web.zoom.us/j/81089906127?pwd=83hivVpSnmnFSVp71pwbEHTuNzj6qK.1

description

As robots increasingly become part of our daily lives, the key challenge is to ensure that they are safe, responsive, and constantly improving while physically interacting with us on the short or the long term. To address this, our workshop discusses three main questions. First, how can we harness innovations in robot design to make robots inherently safe for physical human-robot interaction (pHRI)? This involves the exploration of technologies like soft robotics and haptics to ensure security and comfort in close human contact. Second, how can robots instantly and appropriately react during real-time human interactions? This can be achieved by integrating shared control strategies to make robots adjust their actions to ensure seamless collaboration. Third, how can robots learn from every interaction to be better partners and earn our trust? This encompasses robot learning algorithms and considers how human modeling and feedback can improve these robot partners over time.

The main objectives of this workshop are:

1) To investigate how each of these areas contributes to the improvement of pHRI.

2) To discuss strategies that effectively combine these approaches for a holistic progress in the field.

speakers

Alexis Block

Assistant ProfessorCase Western Reserve University

Title: Design Guidelines for Effective Social-Physical Human-Robot Interaction

Abstract: Hugs are one of the first forms of contact and affection humans experience. Receiving a hug is one of the best ways to feel socially supported, and the lack of social touch can have detrimental effects on an individual's well-being. However, hugs are complex affective interactions that are easy to get wrong because they need to adapt to the approach, height, body shape, and preferences of the hugging partner, and they often include intra-hug gestures like squeezes. We created HuggieBot, a hugging robot, to better understand the intricacies of close social-physical human-robot interaction and as a stepping stone to providing emotional support. Through the iterative design process of creating HuggieBot, we developed 11 tenets of robotic hugging, which ensure a robot can provide its partner with a high-quality embrace. These guidelines can be abstracted to other pHRI applications to enable new possibilities.

Allison Okamura

ProfessorStanford University

Title: Distributed stiffness wearable haptic displays for physical human-robot interaction

Abstract: Haptic devices typically rely on rigid actuators and bulky power supply systems, limiting wearability. Soft materials improve comfort, but careful distribution of stiffness is required to ground actuation forces and enable load transfer to the skin. We present an approach in which soft, wearable, knit textiles with embedded pneumatic actuators enable programmable haptic display. By integrating pneumatic actuators within high- and low-stiffness machine-knit layers, each actuator can transmit 40 N in force with a bandwidth of 14.5 Hz. We demonstrate the concept with an adjustable sleeve for the forearm coupled to an untethered pneumatic control system that conveys a diverse array of social touch signals. We assessed the sleeve’s performance for discriminative and affective touch in a three-part user study and compared our results to those of prior electromagnetically actuated approaches. Our sleeve improves touch localization compared to vibrotactile stimulation and communicates social touch cues with fewer actuators than pneumatic textiles that do not invoke distributed stiffness. The sleeve resulted in similar recognition of social touch gestures compared to a voice-coil array, while in a more portable and comfortable form factor. We propose that this approach will enable more capable, information-rich physical human-robot interaction.

Mehdi Benallegue

Permanent ResearcherNational Institute of Advanced Industrial Science and Technology (AIST)

Title: Staying in Touch: Controlling Human-Robot Interactions with Sustained Contact

Abstract: Current control strategies for human-robot interaction often lack consideration of sustained physical contact, where not only the duration of the contact is long but where the interaction forces and their locations can vary substantially throughout the interaction. This presentation highlights the inherent challenges for ensuring safety guarantees and the solutions that we needed to provide this problem. Additionally, we will raise questions on how to produce communication through interaction forces. To explore this question, we propose to revisit minimum jerk trajectories, known for their motion legibility, and investigate their applicability to sustained contact situations.

Ayse Kucukyilmaz

Associate ProfessorUniversity of Nottingham

Title: Robots Learning from Human Touch: From Conflicts to Collaboration

Abstract: Haptics offers a crucial avenue for human communication during intricate physical tasks. However, despite its utmost importance in physical interactions, the touch modality is under-represented when developing intelligent systems. In this talk, I will discuss our research on haptic shared control, and illustrate examples, where haptic data inform proactive robot partners, which can take on dynamic levels of control, i.e. roles, during collaboration. The talk will demonstrate how physical conflicts and failures can be leveraged in human-robot collaborative work, especially in cases of close physical contact.

Luka Peternel

Assistant ProfessorDelft University of Technology

Title: Design of control methods and interfaces for mutual adaptation in human-robot co-manipulation

Abstract: The talk will present several control methods and interfaces for human-robot co-manipulation that enable robot adaptation to human partners. The adaptation process incorporates shared control based machine learning, human modelling, and real-time measurements to track and improve various metrics, such as task performance, human muscle fatigue, joint torques and arm manipulability. The first part of the talk will focus on the application of co-manipulation for ergonomic physical human-robot collaboration in various practical tasks (e.g., collaborative sawing, polishing, valve turning, assembly, exoskeleton assistance, etc.). The second part will examine teleoperation, where co-manipulation pertains to the remote robot being commanded by a human operator. Here, we will focus on the design of various teleimpedance interfaces and control methods to be used for different applications ranging from manufacturing to remote elderly care.

Dana Kulić

ProfessorMonash University

Title: Learning for Human-Robot Interaction: The Role of Simulations

Abstract: TBD

Harold Soh

Assistant ProfessorNational University of Singapore

Title: Generating Safe Trajectories with Diffusion Models

Abstract: Recent advances in diffusion-based planning models have greatly enhanced robot behavior generation for a variety of tasks. Despite these improvements, adapting these models to new objectives or deployment constraints remains a challenge, often resulting in unsafe behaviors. In this talk, we present a new diffusion-based framework designed to address both safety and temporal constraints specified using finite linear temporal logic (LTLf). Our approach modifies the reverse diffusion process through targeted guidance, allowing for the generation of trajectories that meet LTLf specifications without the need for expert demonstrations for each new instruction. We will also discuss the current challenges in creating safe and compliant trajectories and explore opportunities for developing more reliable and trustworthy robot systems.