No two robots are exactly the same-even for a given model of robot, different units will require slightly different controllers. Furthermore, because robots change and degrade over time, a controller will need to change over time to remain optimal. This paper leverages lifelong learning in order to learn controllers for different robots. In particular, we show that by learning a set of control policies over robots with different (unknown) motion models, we can quickly adapt to changes in the robot, or learn a controller for a new robot with a unique set of disturbances. Furthermore, the approach is completely model-free, allowing us to apply this method to robots that have not, or cannot, be fully modeled.

The shared online calendar is the de facto standard for event organisation and management in the modern office environment. It is also a potentially valuable source of context, provided the calendar event data represent an accurate account of 'real-world' events. However, as we show through a field study, the calendar does not represent reality well as genuine events are hidden by a multitude of reminders and 'placeholders', i.e. events that appear in the calendar but do not occur. We show that the calendar's representation of real events can be significantly improved through data fusion with other sources of context, namely social network and location data. Finally, we discuss some of the issues raised during our field study, their significance and how performance could be farther improved.

Most performance studies of AES make traditional space versus time tradeoffs by allowing large lookup tables to accelerate operations that would normally be calculated by the processor. However, AES is a versatile algorithm and can also be optimised for low-memory use in constrained environments. We investigate the possibility of getting the best of both worlds - an application specific hardware and software solution that has a low dependency on memory yet still executes fast enough to consider for use in production systems. The resulting software is attractive in high level design since it allows AES to be more easily deployed as a composable element in larger systems and scale better as processor speed increases.

The determinism of instruction cache performance can be considered a major problem in multimedia devices which hope to maximise their quality of service. If instructions are evicted from the cache by competing blocks of code, the running application will take significantly longer to execute than if the instructions were present. Since it is difficult to predict when this interference will occur the performance of the algorithm at a given point in time is unclear We propose the use of an automatically configured partitioned cache to protect regions of the application code from each other and hence minimise interference. As well as being specialised to the purpose of providing predictable performance, this cache can be specialised to the application being run, rather than for the average case, using simple compiler algorithms.

Differential power analysis (DPA) has become a real-world threat to the security of cryptographic hardware devices such as smart-cards. By using cheap and readily available equipment, attacks can easily compromise algorithms running on these devices in a non-invasive manner. Adding non-determinism to the execution of cryptographic algorithms has been proposed as a defence against these attacks. One way of achieving this non-determinism is to introduce random additional operations to the algorithm which produce noise in the power profile of the device. We describe the addition of a specialised processor pipeline stage which increases the level of potential non-determinism and hence guards against the revelation of secret information.