Research Structure
Our Centre will achieve breakthrough science and technology in robotic vision by addressing five key research programs. Each program operates a portfolio of projects all focused on developing and integrating robotic vision into a viable technology that will enhance our lives and extend our capabilities.
Sensing
Robots that see in all conditions
Understanding
Robots that see and understand
Acting
Robots that see to act, and act to see
Learning
Robots that learn and improve
Technology
Robots that are fast and low cost
Research Management
The Centre’s research projects are each headed by a Chief Investigator or Senior Research Fellow. Each individual project has a clear objective, with milestones and deadlines. A project team comprises a leader, with a staff of Research Fellows (RF), Associate Investigators (AI), PhD researchers, and Chief Investigators (CI). The litmus test for any Centre project is whether it will develop a technology that demonstrates the capability of robotic vision that can be demonstrated by a robot, ensuring it will deliver an outcome that is relevant to the Centre’s vision.
All new projects are established through our Research Committee, which determines the project’s duration, staffing and other resourcing matters. Project leaders represent their project teams in Research Committee meetings, presenting results and updating the committee on progress. The Research Committee’s role is to ensure project progress, resolve resourcing issues and determine when projects should continue or close.
Activity Plan for 2018
CULTURE
We are creating a vibrant, high energy, future-focused, collaborative robotic vision community developing knowledge leaders for both industry and academia
STRATEGIC OBJECTIVES
- Develop the next generation of robotic vision experts through effective recruitment, retention and training
- Ensure the Centre functions as a cohesive organisation of interactive, collaborative and highly effective research teams
KEY TASKS
- Maintain a full complement of Research Fellows
- Aim for 90% retention of PhD enrolments
- Provide knowledge leadership training to all early career researchers to help in their career development
- Hold one annual symposium each year to build Centre culture
- Run one Robotic Vision Summer School (RVSS) each year as an international PhD recruitment event and to train our new PhD researchers. Continue building the international profile of this event
- Hold monthly meetings of the Research Committee and quarterly face to face meetings
- Centre Executive meet monthly via videoconference plus quarterly face to face meetings, held in conjunction with the Research Committee meeting
- Visit international partners and host visits of international researchers.
TRANSFORM
We solve innovation challenges by applying robotic vision technologies to transform the world
STRATEGIC OBJECTIVES
- Demonstrate how research can advance products and services
- Generate downstream investment to take robotic vision technology into industry
- Foster innovation, entrepreneurship and new enterprises to advance robotic vision
KEY TASKS
- Engage Centre’s Advisory Board at least twice yearly
- Provide knowledge leadership training to all early career researchers to help in their career development
- Present on robotic vision at key industry events
- Launch robotic vision roadmap for Australia.
SCIENCE
We are leading the world in transformational research in the new field of robotic vision
STRATEGIC OBJECTIVES
- Deliver internationally recognised research in robotic vision
- Create and implement projects based on collaboration and innovation that enhance research outcomes
KEY TASKS
RESEARCH PROJECT: MANIPULATION AND VISION
- Achieve continuous integration of visual feedback during grasping in table-top settings. Involving machine learning but simple end effectors (pincher or suction), ie. limited object generality (around 30 types)
- Utilise force-driven interaction for deformable objects on simple gripper
- Assess Reinforcement Learning + Heuristic learning of how to interact with novel objects.
RESEARCH PROJECT: ROBOTIC VISION VIRTUAL EVALUATION TESTBED
- Object detection and semantic segmentation in open-set conditions
- Uncertainty estimation
- Active learning, continuous learning.
RESEARCH PROJECT: ROBOTIC VISION DEPLOYMENT AND EVALUATION
- Implement a systematic study to describe the discrepancy between the current dataset-based evaluation metrics of learning-based predictive models and the true performance of these models onboard real robots. This includes testing state-of-the-art classification, detection and segmentation models onboard robots in realistic settings
- Design methodologies to quantify the performance of the above systems in a variety of real-world environments.
RESEARCH PROJECT: LEARNING
- Use self-supervised learning of 3D models from 2D, 2.5D images
- Achieve uncertainty in 0-shot, open-set and few-shot learning (version 1 of dataset)
- Develop weakly supervised instance segmentation
- Trial methods to compress/simplify Deep Learning models to run in embedded system.
RESEARCH PROJECT: ROBOTS, HUMANS AND ACTION
- Achieve state-of-the-art results on recognising human activities on public benchmarks
- Develop models that compose complex activities in terms of discrete events and short-term actions
- Ability to track object and agents (humans) across time to link activities.
RESEARCH PROJECT: SCENE UNDERSTANDING
- Robustify and properly integrate segmentation with object-based mapping to allow for previously unseen objects
- Incorporate physical (static) relations (supporting, inside, etc)
- Extend static scene understanding to dynamic objects including people (coarse)
- Demonstrate CNN-based single-view depth and pose estimation integrated into full localisation and mapping capability.
RESEARCH PROJECT: VISION AND LANGUAGE
- Grounded visual conversation: Agent trained to participate in a grounded conversation by incorporating progress made towards a fixed task in the reward
- Learning to learn to talk: Use meta-learning tech to train an agent to learn to exploit the capabilities it has and the data available to it. Similar to the Neural Turing Machine but for robots.
RESEARCH PROJECT: FAST VISION-BASED NAVIGATION IN UNSTRUCTURED ENVIRONMENT
- Adaptive perception for visual navigation in unstructured environments
- Techniques for integration of free space affordance (learning based methods) with optical flow to enhance control capabilities.
INTEGRATE
We are bringing the disciplines of robotics and computer vision together to create new robotic vision technologies
STRATEGIC OBJECTIVES
- Connect research organisations, governments, industry and the private sector to build critical mass in robotic vision
- Lead robotic vision in Australia and overseas
KEY TASKS
- Organise robotic vision workshops at key international conferences
- Hold an annual robotic vision summer school targeted at international attendees
- Continue to build network of industry contacts and maintain a customer relationship management system.
ENGAGE
We engage with people about the potential of robotic vision technologies by developing accessible robotic vision resources
STRATEGIC OBJECTIVES
- Identify and engage with key stakeholders on the potential applications of robotic vision
- Establish vibrant national & international robotic vision communities
- Increase robotic vision educational opportunities
KEY TASKS
- Promote our Robotic Vision resources hub
- Promote the Robot Academy
- Host visits and tours of robotic vision facilities by government, industry and the community (including school groups)
- Include popular science articles on Centre work in targeted publications
- Create 4 x media releases by partners related to robotic vision
- Create 4 x public lectures on robotic vision.