Robotic control software is typically modular, featuring various subsystems that solve specific problems, including perception, navigation or motor control.⁵ Historically, these modules have been hand-coded, incorporating physical models of the robot’s hardware and environment, and decision trees telling it what to do in different situations.

Future Horizons:

×××

5-yearhorizon

Mobile robots conscious of their own intent become commonplace

Mobile robots designed to operate in real-world environments become almost entirely AI-powered. Locomotion in diverse environments becomes a solved problem, greatly expanding application areas. LLMs make it possible for non-experts to direct robots through simple natural-language commands.

10-yearhorizon

End-to-end AI replaces modular systems

End-to-end approaches to robotic control replace modular architectures for the vast majority of applications. The most advanced robots start to exhibit sophisticated embodied intelligence, allowing them to undertake increasingly complex tasks and rapidly adapt to new challenges. Breakthroughs in interpretable AI as well as social acceptance of the technology allay some of the safety concerns around widespread robotic deployment.

25-yearhorizon

Self-conscious robots become lifelong learners

As the deployment of robotic systems accelerates, autonomous embodied AI systems are capable of complex social interactions, empathy and moral reasoning. Breakthroughs in lifelong learning make it possible for robots to learn directly from their own experience rather than rely on generic pre-trained models.

This can achieve impressive results. Boston Dynamic’s agile humanoid robots use this approach,⁶ for instance — but typically only for highly constrained tasks. Advances in AI mean these modules are increasingly being replaced by more flexible, learned models. Deep learning already powers Simultaneous Localization and Mapping (SLAM) in autonomous vehicles and drones,⁷ and helps sorting robots pick out and manipulate objects.⁸Progress is being made in applying AI to locomotion and planning too.⁹

Transformers — the algorithms behind large language models (LLMs) — promise to greatly expand AI’s use in robotics.¹⁰ LLMs themselves provide a more natural way to interface with robots via language.¹¹ But transformers also power new multimodal models that can work with multiple data types.¹²

At the cutting edge, this is driving a shift from modular architectures to “end-to-end” models that simultaneously solve perception, control and planning.¹³ The ability to draw correlations between different data sources, including language, vision and motion, could allow robots to develop a deeper understanding of the physical world and their place in it — something referred to as embodied intelligence.¹⁴,¹⁵,¹⁶ Through interactions with the real world, robots also have the potential to overcome AI’s “symbol grounding problem”¹⁷, able to link the learned word or “symbol” of an object, such as “cup”, to the object’s real-world context. In turn, embodiment may allow AI to develop a sense of “self”¹⁸¹⁹, although a comprehensive theoretical understanding of consciousness and its emergence from physical and social interactions remains a challenge.

Moreover, current models are data- and power-hungry and incapable of learning on the fly. Their decision-making is also inscrutable, raising safety concerns. The seamless integration of sensory, motor and cognitive functions will be required to enable real-time learning and adaptation, and breakthroughs in lifelong learning and interpretability will be crucial in the long run.²⁰

Robotic software - Anticipation Scores

The Anticipation Potential of a research field is determined by the capacity for impactful action in the present, considering possible future transformative breakthroughs in a field over a 25-year outlook. A field with a high Anticipation Potential, therefore, combines the potential range of future transformative possibilities engendered by a research area with a wide field of opportunities for action in the present. We asked researchers in the field to anticipate:

The uncertainty related to future science breakthroughs in the field
The transformative effect anticipated breakthroughs may have on research and society
The scope for action in the present in relation to anticipated breakthroughs.

This chart represents a summary of their responses to each of these elements, which when combined, provide the Anticipation Potential for the topic. See methodology for more information.