On 20/10/06, Pei Wang <[EMAIL PROTECTED]> wrote:
Part of the reason for recent successes in robotics is simply about computing power, enabling sensor data to be processed in something approaching real time. Part of it has to do with general information technology infrastructure, such as GPS receivers, inertial sensors, digital cameras, small PC motherboards and so on. All these things were available years ago, but at much higher cost and complexity making it harder to integrate into a single system. A decade ago attaching a camera to a robot was a big deal and involved what at the time was a lot of processing power. Now it's no big deal to have multiple cameras attached with images being processed at video speeds.
But the main advances in robotics at present are coming from improved algorithms for localisation and recognition of objects. Increasingly we're seeing the use of kalman filters, particle filters, more sophisticated sensor models and fusion between multiple sensor types, resulting in more robust performance. I think we're now at a crossing point where robotics is moving from a situation where the navigation ability was previously a bit flaky and would usually get lost after a period of time to a situation where navigation is effectively a solved problem and the performance is highly reliable.
Another interesting development is the rise of the use of invariant feature detection algorithms together with geometric hashing for some kinds of object recognition. The most notable successes to date have been using David Lowe's SIFT method, which I think bears some resemblence to earlier methods such as Moravec's interest operator. To an extent these are just old algorithms developed in the 1980s enjoying a new lease of life within a more favourable computational environment.
It should also be noted that recent successes have been based upon probablistic methods. In the past I think it wasn't realised how useful it is to consider observed features within the environment in terms of probability distributions. This especially applies in my own area of stereo vision, where incorrect assumptions about triangulated features being single points in space have dominated for many years.
Because of this, I'm not sure how long robotics can keep
its recent improving rate without major progress in AI in general.
I wonder if there is anyone in this list who has been actually working
in the field of robotics, and I would be very interested in learning
the causes of the recent development.
Part of the reason for recent successes in robotics is simply about computing power, enabling sensor data to be processed in something approaching real time. Part of it has to do with general information technology infrastructure, such as GPS receivers, inertial sensors, digital cameras, small PC motherboards and so on. All these things were available years ago, but at much higher cost and complexity making it harder to integrate into a single system. A decade ago attaching a camera to a robot was a big deal and involved what at the time was a lot of processing power. Now it's no big deal to have multiple cameras attached with images being processed at video speeds.
But the main advances in robotics at present are coming from improved algorithms for localisation and recognition of objects. Increasingly we're seeing the use of kalman filters, particle filters, more sophisticated sensor models and fusion between multiple sensor types, resulting in more robust performance. I think we're now at a crossing point where robotics is moving from a situation where the navigation ability was previously a bit flaky and would usually get lost after a period of time to a situation where navigation is effectively a solved problem and the performance is highly reliable.
Another interesting development is the rise of the use of invariant feature detection algorithms together with geometric hashing for some kinds of object recognition. The most notable successes to date have been using David Lowe's SIFT method, which I think bears some resemblence to earlier methods such as Moravec's interest operator. To an extent these are just old algorithms developed in the 1980s enjoying a new lease of life within a more favourable computational environment.
It should also be noted that recent successes have been based upon probablistic methods. In the past I think it wasn't realised how useful it is to consider observed features within the environment in terms of probability distributions. This especially applies in my own area of stereo vision, where incorrect assumptions about triangulated features being single points in space have dominated for many years.
I'm old enough to have lived through the hopeful monsters of AI, characterised by brief periods of early success followed by longer periods of stagnation. For most of the time over which I've been interested in AI (more than 15 years) lack of computing power has been a major barrier, but now that's no longer the main issue and instead it's all about algorithms. I completely agree that there won't be much further progress in robotics unless there are also corresponding developments in AI algorithms and architectures.
This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
