Today’s state-of-the-art AI is incredibly data-hungry, which presents significant challenges for robotics applications. Most recent advances have come in domains like computer vision and natural-language processing, which can take advantage of the reams of image and text data on the internet. But real-world robotics data is much harder to come by, constraining the ability to train sophisticated models.

Future Horizons:

×××

5-yearhorizon

Data remains a bottleneck

Data remains the biggest bottleneck on robotic advances, despite the increasing availability of open robotics data. This spurs the development of more data-efficient training approaches. Ambitious governments start to set up national robotics centres with large numbers of robots that can carry out more ambitious experiments and collect more data.

10-yearhorizon

A virtuous circle eases data woes

Widespread deployment of robots leads to a steadily increasing deluge of robotics data. This helps accelerate advances, providing a further boost to deployment and creating a virtuous circle of progress. The robotics companies that build the devices are the greatest beneficiaries.

25-yearhorizon

The data crunch is over

Data access ceases to be a significant problem due to the vast amounts being produced by widely deployed robots and more data-efficient training approaches. Data is no longer a differentiator, encouraging private companies to open up their datasets, providing a boost to academic research.

Transformer-based AI presents both opportunities and fresh problems. Transfer learning — the ability to take a model trained on one kind of robot and deploy it on another — has been a long-standing challenge.³¹ Even small changes in environment or robot configuration can throw models off, meaning robots have to be individually trained. Transformers are able to train on data from multiple robots to create more general policies that work across varied embodiments and environments.³². However, training one of these models requires colossal amounts of data.

One potential workaround involves training models in simulations before porting them over to real-world robots.³³ But many of the tasks planned for robots, such as handling soft and delicate objects, are incredibly hard to simulate.³⁴ Designing virtual worlds is also complicated and expensive, requiring sustained effort from multidisciplinary teams.³⁵

Pooling data-collection efforts will be crucial going forward, and there are promising efforts to create massive open robotics datasets.³⁶ There will also be a growing focus on more data-efficient training approaches.³⁷ Data shortages are likely to ease as more robots are deployed, though this will benefit the private companies that build them more than academic researchers.

Data - 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.