Adaptive assistive robotics: a framework for triadic collaboration between humans and robots

Abstract

Robots and other assistive technologies have a huge potential to help society in domains ranging from factory work to healthcare. However, safe and effective control of robotic agents in these environments is complex, especially when it involves close interactions and multiple actors. We propose an effective framework for optimizing the behaviour of robots and complementary assistive technologies in systems comprising a mix of human and technological agents with numerous high-level goals. The framework uses a combination of detailed biomechanical modelling and weighted multi-objective optimization to allow for the fine tuning of robot behaviours depending on the specification of the task at hand. We illustrate our framework via two case studies across assisted living and rehabilitation scenarios, and conduct simulations and experiments of triadic collaboration in practice. Our results indicate a marked benefit to the triadic approach, showing the potential to improve outcome measures for human agents in robot-assisted tasks.

Keywords: ergonomics; optimal control; optimization.

Figure 1.

Triadic collaboration lies at the intersection of human–robot teaming and pHRI.

Figure 2.

A general illustration of triadic collaboration. In the general case, this comprises a mixture of agents, of which at least one is human and at least one is technological, with some level of physical interaction between agents. An example, as depicted, is a nurse and robot collaborating to help a patient perform a sit-to-stand.

Figure 3.

(a) A research participant wearing a lower-body exoskeleton, taking part in an investigation of sit-to-stand biomechanics. (b,c) Snapshots from a reconstruction of the sit-to-stand motion. Musculoskeletal models, built in OpenSim [30], account for the coupling between human and exoskeleton, and allow for detailed analysis of the behaviour of human joints and muscles.

Figure 4.

A schematic outlining the relationship between the key components of the triadic collaboration framework. This can be interpreted as a control diagram showing inputs to and outputs from each component of the framework at time t. The formulations discussed in this work refer to the optimization component, shown in green.

Figure 5.

A schematic illustrating the triadic collaboration framework instantiated in an assisted living scenario. Here, a human agent such as a nurse or carer (the assistance provider) is providing physical assistance to a patient (the assistance seeker) with additional support from a technological agent (e.g. an exoskeleton). The assistance seeker has direct physical interaction with both the technology and the assistance provider, while the assistance provider interacts with the technology only in a supervisory fashion.

Figure 6.

Snapshots of an example triadic collaboration work task involving two human agents and an exoskeleton. One human agent bears the brunt of the load, with the assistance of an exoskeleton providing ergonomic support. Meanwhile, the second human agent carries out the finer manipulation.

Figure 7.

Snapshots of a simulated assisted sit-to-stand transition. A patient (left) is assisted in completing a sit-to-stand by a combination of a human carer (right) and an exoskeleton (highlighted in yellow). The red plane models a seat, while blue spheres represent contact geometries.

Figure 8.

(a) The assistive torques generated by the APO in simulations 3–6. (b) The lumbar loading and stability costs for each simulated test case. Lower costs indicate better performance.

Figure 9.

Intermediate snapshots from simulations 3 (a), 4 (b) and 5 (c) showing the agent configurations after 0.5 s of the sit-to-stand transfer. Note in particular the more acute angle of the assistance provider’s back joint in Simulation 3, in which lumbar loading is not optimized.

Figure 10.

A schematic illustrating the triadic collaboration framework instantiated in a specific rehabilitation scenario. Here, a human agent carries out a prescribed motion with assistance from two technological agents—an exoskeleton and FES electrodes.

Figure 11.

Snapshots of a simulated assisted swing leg motion. A patient is assisted in completing the swing leg motion by a combination of exoskeleton assistance and electrical stimulation. During the initial swing (a), the gluteus muscle is stimulated, during the mid-swing (b) the hamstring muscles are stimulated and during the terminal swing (c) the vasti muscles are stimulated. The stimulated muscles are presented in red and the non-stimulated muscles are presented in blue.

Figure 12.

(a) A subject undergoing familiarization with the FES electrodes. (b) A subject undergoing the trajectory tracking task with the H3 in transparent mode, to obtain the baseline human intention.

Figure 13.

A comparison between three different rehabilitative interventions; exoskeleton-only assistance, FES-only assistance, and hybrid exoskeleton-FES assistance. For each of the three interventions, the estimated tracking error, assistance and muscle fatigue are compared between the case where baseline controller parameters are used and the case where the parameters are optimized for triadic collaboration.