The characteristic feature that turns a machine into a robot is the ability to take autonomous action. This may be through the manipulation of objects of locomotion through the world. Action encompasses both the knowledge required for a robot to interact with its environment, and the actuators and control systems required to enable a robot to move and take action.

When robots take action, it is usually to perform a specific task or set of tasks. There is still a long way to having the technology required for multi-purpose robots, for example a vacuum-cleaning robot that could also pick up clothes. The drivers to have complicated multi-purpose robotic systems are not yet apparent but may emerge in the future. The breakthroughs required to improve the ‘action’ capability of robots include clever use of physical principles, improved actuators and advanced materials to improve robot movement strength, endurance and range (section 11.6), and deeper integration of action, perception and cognition [AAS18]. This section deals with the former, while section 11.6 deals with advances in actuators and control.

For robots to navigate and explore involves path planning, obstacle avoidance, localisation and mapping, which involves: advanced simultaneous localisation and mapping (SLAM) techniques, semantic understanding (section 11.2), cooperation (section 11.3), learning (section 11.4), and robustness (section 11.6). Robots need to be able to learn, forget, and associate memories of scene content, to gain an in-depth (semantic) understanding of their environment, to reason and discover and distinguish new objects through learning, and to evolve via continuous learning [SM18]. They also need to be able to adapt, learn, and recover after mistakes, to make and recognise new discoveries, and have the physical robustness to withstand harsh, changeable environments, rough handling, and complex manipulation, and to self-repair when required [SM18].

Action is an integral function of a robot with its surroundings, human users, and other robots, and is integrated with perception and cognition (section 11.2). Robots must act in increasingly complex environments, often under unknown and/or extreme conditions with limited, or no, communication. For example:

Robots in tunnels or mines must cope with rough terrain, narrow passageways, and degraded perception.

Robots undertaking nuclear decommissioning must withstand radiation and restricted access.

Robots used to construct and assemble infrastructure must be resistant to dirt, dust, chemicals and large impact forces [SM18].

The long latency and low bandwidths of communication in these environments slows down robot action and can also pose a risk to the robot’s survival. Australia is particularly impacted by low bandwidth – high latency challenges due to our immense geography and poor network coverage. For this reason, the technologies for robots to be able to act autonomously, despite these challenges, is imperative.

For robots to manipulate objects requires dexterity. Improvements in manipulation are required for robots involved in almost all service robotics applications [US16]. Grasping, in-hand object manipulation, and the execution of complex and intricate tasks, requires mechanism design, materials, planning and perception and involves the collection of a range of information, such as tactile feedback, object shape, contact locations and centre of mass. Manipulation also requires a clear control algorithm to effectively execute a manipulation task [SM18]. Soft compliant skin with inbuilt tactile sensing, where a robot senses the degree of force required to complete a manipulation task and to reliably grasp objects, remains at immature levels and does not come close to human capabilities. There is still five to 10 years of research and development required to progress the state-of-the-art to enable robots to do simple household tasks, such as stacking a dishwasher.