AI for Robotics (ELEN90095)

AIMS:  

The key focus of this subject is the foundations of robotic systems that use software to move autonomously through their environment. This subject focus on the software & algorithms that enable the robot to perform tasks autonomously. Hence, this subject focused on artificial intelligence (AI) software & algorithms for robotics. The first main aim of the subject is to provide a foundation of the feedback loop that is core to all AI-enabled robots, namely: sensors measure the world around the robot; AI algorithms decide what action to take; the robot enacts that action by moving its joint or wheels; and the loop repeats endlessly. The second main aim of the subject is to provide implementation experience with cutting edge AI algorithm applicable to consumer and industrial robotics, where we consider both model-based method and reinforcement-learning methods. 

INDICATIVE CONTENT:   

Topics covered are at the intersection of automatic control and artificial intelligence, including:  

• Cyber-physical feedback system formulation, such as: black-box and grey-box modelling, stability and robustness safety requirements, hierarchical and network control architectures.  
• Safety and convergence guarantees for model-based methods, such as: learning models from data; adaptive control schemes; stability and robustness of PID and MPC control approaches. 
• Connections between optimal control and reinforcement learning formulations for robotics.  
• Reinforcement learning for robotics, such as: actor-critic methods, on-policy versus off-policy learning, sample efficiency, transferring simulation-based learning to real-world robots

Intended learning outcomes

On completion of this subject, students should be able to:

• Construct and critique systems-level formulations and architectures for physical systems to perform tasks, using tools such as block diagrams, systems thinking, and Markov decision processes.
• Analyse and implement model-based and learning-based methods for autonomy of physical engineering systems, with continuous spaces, under the control of discrete-time policies.
• Appraise the boundaries of, and combine the strengths of, multiple decision-making algorithms for implementation on real-world systems, including model-based and learning-based algorithms.
• Employ reflective practise for the purpose of developing their interdisciplinary perspective for the knowledge bases, methodologies, and norms from the engineering, computer science, and wider domains.
• Communicate and collaborate on AI and decision-making projects through formats such as: technical reports, presentations, and informational videos, and reflection

Generic skills

• Ability to identify, formulate, analyse and solve complex problems using a systems approach
• Capacity for independent critical thought, rational inquiry, creativity, and innovation
• Ability to communicate effectively through verbal and written means with a variety of target audiences, including with one's direct team, other engineers and professions, and with the community at large
• Ability to function effectively in multi-disciplinary and multi-cultural teams of various sizes, with the capacity to be an effective leader, manager, and team member as appropriate to the context
• Ability to scope and plan project work and manage competing demands on time to deliver a substantial outcome, including any self-directed learning necessary