- Core temperature is too high
- Required Action = lower temperature
- What happened previously that led to a lower temperature?
- Open a window (but only if outside temperature is lower)
- Stop processing unnecessary actions to let the hardware cool down
This needs some recording of what the robot does and the outcome.
- e.g. when i opened the window the core temperature went – Sometime other conditions need to be assessed. Opening the window when the temperature outside is lower causes core temperature to go down, however opening the window when the temperature outside is higher would cause the core temperature to go up
- Often many things will be happening at once so it may not be clear what directly causes the temperature to go down
- If the core temperature always goes down when I open the window, then it’s a high probability that it will work again
- If the core temperature only went down once when I opened the window (out of 100 times opened) then it’s unlikely that opening the window again will cause the core temperature to go down.
Unstructured Big Data principals might be needed here. I need to investigate how to best model these types of information / relationships and then query them.
This is about tracking “Cause and effect”. E.g. Performing intense activity “moving at full speed” causes battery levels to drop. AI needs to pattern match on the data to evaluate the cause and effect. Was it the intense activity or something else?
AI can use a history of these cause and effects to predict future effects based on chosen activities.
Also, changes to the “System” need to be monitored, such as installing a bigger battery means that the system can perform that intense activity for longer before the battery runs down.
My anticipation for controlling the robot is to use multiple Raspberry Pi’s that all serve a specific role.
There is no reason it has to be multiple Pi’s, but since they are low powered they probably can’t do everything on a single unit. We can better think of the different components as modules that can be run on the Pi with the most available processing capability… maybe a control unit can spin up the services as required on a suitable Pi unit.
Some functions will “need” to be run on a certain Pi, for example a Pi connected to a sensor is the only Pi that can read the values from that sensor. Maybe there will need to be some specific uni-role Pis and then some more generic Pi’s for “decision making” and “information processing”.
I envisage the following Pis:
- Vision – Web Cam, Radar, Kinnect (this will need something more powerful than a Pi)
- Sound processing
- Sensing and positioning sounds
- Processing and identifying the sounds
- Raising Alerts based on the sound (the following actions will be managed from a Control Unit that picks up the “emergency alert”, not this sound processing Unit)
- If a large bang and close by then raise an “emergency” alert and divert power to further identify the sound
- Move the vision sensors to look at it
- Identify if there is a danger
- If not,
- “Remember” that sound profile and that a large bang nearby from a balloon popping is not dangerous
- Carry on with previous task(s)
- If so
- Take precautionary action
- Protect sensitive areas
- Move away
- Establish where to move to
- Is it safer?
- Vocal processing for audio feedback
- Decide what actions to be taken to increase the “happiness quotient”
- Respond to emergency alerts
- Request movement / action from motor controls
- Emotional engine
- Monitor “happiness quotient” and individual factors affecting it
- Raise alerts if “happiness quotient” drops below a certain threshold
- Raise alerts if any individual factor drops below a certain threshold
- Such as power / energy levels drop below a certain level
- E.g. When the battery runs low, raise an emergency alert and the control system should dicate the most important action is to find food (power)
- Sensor input(s)
- Vision / Radar
- Central nervous system
- Spatial Awareness
- Process sensors inputs (vision, sound, touch, etc) to build an image of the world around
- Process all sensory input and fire alerts if something needs attention
- Use spatial awareness to combine inputs and deduce if something needs attention, such as the sound of a car and the sight of it getting close could be impending danger – raise an emergency alert
- Motor control
- Move something forwards / backwards / left / right / around
- Move an arm
When storing and recalling memories it is well known that we don’t remember every minute detail of a situation, instead some of the key attributes of the and then recall will rebuild the situation and fill in the gaps with “intuition” and “inferred knowledge”.
For example, I remember sitting down to breakfast this morning and eating cereal.
I remember I had Kellogg’s Shreddies and sat at the kitchen table facing my fiancé, Amy.
This is the memory, but there is a lot else that comes back when I think about that situation, such as:
- I know Amy was there because I remember her asking me to set up breakfast.
- I only ever “set up breakfast” when Amy is there.
- I must have had milk as I always have milk on my shreddies
- Therefore I must have gone to the fridge to get it – added to the “story” of the original memory and “comes to light” as I develop the story of the situation of me eating my breakfast.
- I was in the new kitchen (I ate breakfast this morning in this memory and I know it’s 2013 and the new kitchen was fitted in 2012), so lots of separate memories of being in the kitchen allow me to form a complete view of what the kitchen looks like. This won’t be “accurate”, but the non-moving items will be in place.
- I know the coffee machine is in it’s place – 100% probability as it’s never been anywhere else
- I know the TV is in the corner
- and probably on as it is normally on when I eat breakfast – 80% accurate
- I was facing Amy, so it must have been a Saturday or Sunday as we only eat breakfast together at the weekend
- It was probably showing Saturday Kitchen as that is normally on at the time I eat breakfast on a saturday
- I don’t have any tea towels in the “image” of the kitchen as they could be anywhere (too low probability so eliminated from the recall image)
- There is a 50-50 chance there were flowers on the table
- I “may” include flowers in the memory if it makes the memory happier (relationship to the emotion engine)
- If I am in a sad mood when recalling the situation then I may “exclude” the flowers to make the memory more applicable to my current negative emotional state
- If I am in a happy mood when recalling the situation then I may “include” the flowers to make the memory more applicable to my current positive emotional state
- This suggests / supports memories are not “reliable” evidence and can be affected by external factors such as current emotional state.
- Think of a solicitor asking a witness “Just how slow was the car travelling?” as opposed to “Just how fast was the car speeding towards you at?” It is proven that these different questions of the same scene will affect the answer quite considerably.
- I remember it is the brown kitchen table that was in front of me
- This wasn’t stored with my original memory of eating shreddies, but inferred from the following previous memories
- There has been a white table in the kitchen
- There has been a brown table in the kitchen
- The white table was put in the kitchen in 2010
- The brown table was put in the kitchen in 2012
- I ate my breakfast in 2013, therefore it must have been the brown table in the kitchen when I ate my breakfast
- This shows that being able too “time-box” situations is important to select related situations on a “Temporal” basis
The longer ago the breakfast memory was the more it “fades” (probability of each supporting piece added drops).. for example if I think back to a breakfast a few years ago, I can’t quite remember if it was in 2011 or 2012 so it may have been the new or old kitchen, so this is a “faded” (low probability) memory and only the very important attributes are clear and trusted (the fact I ate breakfast facing Amy).
- Whether is is the new or old kitchen is more faded as I can’t quite remember the year
- I know it was the weekend as I was facing Amy, this is still clear
- Saturday morning kitchen was on TV as that is nearly always (80% probability) on the TV when I eat breakfast facing Amy
I have been thinking about the role of emotion in decision making for a long time. I have not been influenced by any existing research in the area, but focusing on my own thoughts. A quick Google this morning highlights others have also had these ideas, but only fairly recently (within the last few years) and it’s full role in decision making is still highly disputed in favour of “rational” decision making.
I think this is totally and utterly wrong. I think emotion is the MAJOR factor in decision making and rational is a method of implementing the emotional decision. Sometimes Rational may decide that the emotional decision is not achievable (or too risky – maybe the “happiness reward” is so far negative that it’s not worth the risk) and that a number of smaller, more achievable decisions will give a bigger “happiness reward” and override the emotional decision. Sometimes the “happiness reward” of an emotional decision is so great that the rational engine cannot override it and a “risk” is taken.
Everything we do is in the pursuit of happiness, or rather “perceived happiness”.
I have an interesting line of thought around self-destructive behaviours that are still born out of experience and I think those people still make decisions that are deemed self-destructive in the pursuit of happiness. It is just that their experiences have been interpreted incorrectly and they make bad decisions. For example, self-harming, people will self harm, because having that physical pain temporarily eliminates the mental pain that the person is suffering, so, they interpret that to mean that self-harming will increase their happiness quotient. This is where we need rationality to layer on top of emotional decisions to keep them in check,
This Category (Robotics) is a collection of thoughts and ideas about Cybernetics – to me, this is encompassing Robotics, Artificial Intelligence and Control Mechanisms.
My intention is to build an autonomous robot that can think and make decisions for itself.
After years of contemplating and thinking about different aspects these areas, which were initially born out of an interest in AI from my Applied Computer Science Undergraduate course at Reading University between 1997 and 2001, I have decided to document some of my thoughts and see if I can develop them into a physical being – a robot.
With the development and release of Raspberry Pi, I believe the technology is there, and affordable for a lay-scientist to develop something out of those ideas.
Watch this Category for more of my random musings.