Object handover is a common task arising frequently in many cooperative scenarios. Therefore, it is crucial that robots perform handovers well when working with people. However, determining the proper handover method for an object is a difficult problem since it varies depending on each object’s affordances. Towards enabling effective human-robot cooperation, this thesis contributes a framework that enables robots to automatically determine handover methods for various objects by observing human handovers and object usages.
This thesis first documents a user study conducted to characterize and compare the handover orientations used by humans in different conditions. It puts forth the novel idea of object affordance axes for identifying patterns in handover orientations, and a distance minimizing method for computing mean handover orientation from a set of observations.
Next, this thesis presents an object grouping and classification method based on observed object usage for generalizing learned handover methods to new objects. Until now, a demonstrated method for generalizing handover methods to new object has been lacking. The presented method focuses on a set of action features extracted from the movement patterns and inter-object interactions observed during usage. An experiment demonstrates the effectiveness of the method on grouping objects and then classifying new objects and computing proper handover methods for them.
The described framework for learning and generalizing handover methods is implemented onto a Kawada Industries HRP2V robot, and this thesis also documents the verification experiments. The implementation in this thesis overcomes the robot perception challenge of identifying a held object’s pose at handover by detecting the object at the pre-occluded state and tracking its pose using a sequential Monte Carlo method. Results show that the framework allows robots to learn handover methods from demonstrations and compute proper handover methods for new objects. This is the first demonstrated system capable of automatically learning and generalizing handover methods from observations. Finally, integration into a household service robot application shows how this work this can enhance the capabilities of robots working in the real world by enabling them to work effectively with humans.
Through enabling better human-robot object handovers, this thesis contributes towards improving the interaction between humans and robots, thus, allowing safer, more natural, and more efficient human-robot cooperation.
Handing over objects is a common basic task that arises between people in many cooperative scenarios. On a daily basis, we effortlessly and successfully perform countless unscripted handovers without any explicit communication. However, handing over an object to a person is a challenging task for robotic “hands”, and the resulting interaction is often unnatural. To improve human-robot cooperation, the work described in this thesis has led to the design of a human-inspired handover controller based on analysis and characterization of the haptic interaction during human-to-human object handover.
The first experiment in this thesis documents novel experimental work done to measure the dynamic interaction in human-human handovers. The grip forces and load forces experienced by the giver and the receiver during a handover are examined, and the key features are identified. Based on these experimental results, guidelines for designing human- robot handovers are proposed. Next, this thesis describes a handover controller model that enables robots to hand over objects to people in a safe, efficient, and intuitive manner, and an implementation of the handover controller on a Willow Garage PR2 robot is documented. Finally, a second experiment is presented, which compares various tunings of the novel controller in a user study. Results show that the novel controller yields more efficient and more intuitive robot-to-human handovers when compared to existing handover controllers.
My research focuses on developing communication mechanisms for human-robot manipulation interaction. I enjoy working in detail-oriented and multidisciplinary teams to translate research from HRI and computer vision to advance mission-critical system. Thanks to my multidisciplinary background, I have the capacity of turning a novel idea into a functional prototype.
In recent years, robots have started to migrate from industrial to unstructured human environments, some examples include home robotics, search and rescue robotics, assistive robotics, and service robotics. However, this migration has been at a slow pace and with only a few successes. One key reason is that current robots do not have the capacity to interact well with humans in dynamic environments. Finding natural communication mechanisms that allow humans to interact and collaborate with robots effortlessly is a fundamental research direction to integrate robots into our daily living. In this thesis, we study pointing gestures for cooperative human-robot manipulation tasks in unstructured environments. By interacting with a human, the robot can solve tasks that are too complex for current articial intelligence agents and autonomous control systems. Inspired by human-human manipulation interaction, in particular how humans use pointing and gestures to simplify communication during collaborative manipulation tasks; we developed three novel non-verbal pointing based interfaces for human-robot collaboration. 1) Spatial pointing interface: In this interface, both human and robot are collocated and the communication format is done through gestures. We studied human pointing gesturing in the context of human manipulation and using computer vision; we quantified accuracy and precision of human
pointing in household scenarios. Furthermore, we designed a robot and vision system that can see, interpret and act using a gesture-based language. 2) Assistive vision-based interface: We designed an intuitive 2D image-based interface for upper body disabled persons to manipulate daily household objects through an assistive robotic arm (both human and robot are collocated sharing the same environment). The proposed interface reduces operation complexity by providing different levels of autonomy to the end user. 3) Vision-Force Interface for Path Specication in Tele- Manipulation: This is a remote visual interface that allows a user to specify in an on-line fashion a path constraint to a remote robot. By using the proposed interface, the operator can guide and control a 7-DOF remote robot arm through the desired path using only 2-DOF. We validate each of the proposed interfaces through user studies. The proposed interfaces explore the important direction of letting robots and humans work together and the importance of using a good communication channel/ interface during the interaction. Our research involved the integration of several knowledge areas. In particular, we studied and developed algorithms for vision control, object detection, object grasping, object manipulation and human-robot interaction.
PhD Candidates and Students
Supervisors: Drs. Machiel Van der Loos, Jaimie Borisoff
There are a large number of people all around the world who rely on wheeled mobility assistive devices (WMAD) to perform their daily life activities. The use of WMADs impacts various aspects of peoples’ life including their personal autonomy. In many cases, autonomy – that is, peoples’ choices and controls over what they want to do – is determined by the type of mobility assistive device they are using. Therefore, it is essential to recognize, assess, and address the true autonomy-related needs of mobility device users in the process of assistive device development.
In my research, I’m reviewing the literature to identify the main contributing factors to the autonomy of WMAD users. Next, I compare the design and performance characteristics of existing WMADs across these factors. This knowledge provides an insight into the existing gap between the users’ needs and what is available to them. To address this gap, I plan to establish an autonomy-based framework for mobility assistive technology development. Use of this framework could lead to the design and development of mobility assistive devices that provide a more balanced sense of autonomy to the users
Supervisors: Drs. Machiel Van der Loos, Jaimie Borisoff
Powered lower limb exoskeletons (LLEs) are wearable robotic aids that provide mobility assistance for people with mobility impairments. Despite their advanced design, LLEs are still far from being effective assistive devices that can be used to perform activities of daily living. The main challenge in the operation of a LLE is to ensure that balance is maintained. However, maintaining an upright stance is not always achievable and regardless of the quality of user skill and training, inevitably falls will occur. Currently, there is no control strategy developed or implemented in LLEs that help reduce the user’s risk of injury in the case of an unexpected fall.
In this thesis, an optimization methodology was developed and used to create a safer strategy for exoskeletons falling backwards in a simulation environment. Due to the data available regarding the biomechanics of human falls, the optimization methodology was first developed to study falls with simulation parameters characteristic of healthy people. The resulting optimal fall strategy in this study had similar kinematic and dynamic characteristics to the findings of previous studies on human falls. Rapid knee flexion at the onset of the fall, and knee extension prior to ground contact are examples of these characteristics. Following this, the optimization methodology was extended to include the characteristics of an exoskeleton. The results revealed that the hip impact velocity was reduced by 58% when the optimal fall strategy was employed compared to the case where the exoskeleton fell with locked joints. It was also shown that in both cases of optimal human and human-exoskeleton falls, the models contacted the ground with an upright trunk with a near-zero trunk angular velocity to avoid head impact. These results achieved the thesis goal of developing an effective safe fall control strategy. This strategy was then implemented in a prototype exoskeleton test device. The experimental results validated the simulation outcomes and support the feasibility of implementing this control strategy. Future studies are needed to further examine the effectiveness of applying this strategy in an actual LLE.
The first prototype consisted of a model of a triple-link inverted pendulum and a control system, designed and fabricated by an undergraduate Engineering Physics student team at UBC. The mechanical test setup characterized a half-plane and half-scale model of a human body. Three joints of the triple-link inverted pendulum replicated the motion of the hip, knee, and ankle joints. Similar to the three-link model of a human fall, the hip and knee joints of the inverted pendulum were actuated and the ankle was a passive joint. The hip and knee joint angles were read through the actuator’s encoder and the ankle joint angle was read by a potentiometer that was installed at the joint. The controller was programmed to start the safe fall control strategy once the ankle angle passed beyond a specified angle. Therefore, the ankle angle sensor was constantly monitored subsequent to the initialization of the hip and knee joints. When the ankle angle exceeded the specified limit, the position control strategy was activated to control the hip and knee joint angles throughout the fall.
Large deviations were observed between the experimental and optimal values of the joints angular velocity throughout the fall duration. This is mainly due to hardware and software limitations of this prototype.
To address the abovementioned issues and to further improve controller performance a second prototype was built by an undergraduate Mechanical Engineering student team at BCIT. The second prototype includes a scaled and adjustable exoskeleton with the same actuation setup as the first prototype and will execute the fall routine, as well as a release mechanism to zero the system and initiate the fall. Currently, we are working on implementing the developed safe fall strategy in this prototype.