# Full Text: BevCyc - posit on primitive drivers of creatures > Extracted from `bevcyc.pdf` --- ## Page 1 BevCYC GIST BehavioralCycle / CYC / “ from epigenetic-sequence / epigenetic-cycle” transcribed onto ROS-like / gymnasium like Nodes-of-action-sequence; but in actual essence; is just an empty node-scaffold To fill with action-context-pairs that is designed so that they are decomposable Onto the lowest level (which is muscle tension %) “ BevCYC is a framework that implements a decomposable Chain of Node ( N – N – N ) Representing an Action Sequence / Concept Sequence That is based on the Epigenetic Cycle ideations. (#p , #p ) (commonly known changing / semi temporary / semi permanent ; biological instruction set) While currently structured in such way (empty nodes for pairs) (in a 3-3-3 minimum or more) , each sequence can be otherwise interpreted as a “state” – “action” – “state” triplet (more commonly known nomenclature seem to imply a distinction) . In the mind of the cat; (in this framework) “state” – “action” –“state” is actually a “state” – “state” – “state” (hunger.) – (eating. / hunting. / eat-ate-ing) – (ate.) The distinction between the states and the action is deliberately blurred Because they are prepped in such way where ## Page 2 They (hopefully) eventually just represent nearest node-of-activation That can be both represent the action and the condition. (or mayhaps more easily both referred to as “concepts”) Through hierarchal decomposition; these ultimately resolve onto a muscle Tension % On practice; these nodes are usually taken (copied) Onto a learning template; for a adjacent Agent / LLM to then work on; To update and then re-copy / re transcribe On differing time spans; / intended time-length . As in; this implies that in biology; the same molecule is “copied” For base reference when practicing the action for the first time; . Also such is kept; as a “human-sense” initial copy of what that action should be compared-to (imagine instinctively know what is “that is very primitive / very human”) Or otherwise (“this is CRANE KUNGFU, embody CRANE! O_O >:OO”) (unclear instructions that in the chemical level; is actually faster to transcribe) (due to the same-ness of both the CYC / learned CYC / state & action) . Also such is parsed / edited onto a learned context-specific version (copy) With its own % tension range (of the muscles) We will try to describe how a typical chain is copied and liberally modified For % application and etc (these are probably done in such way in biology for ease/recoverability) ## Page 3 Unfurling On practice Upon unfurling; they become 3 different (shortening) node of interjection / injection of decision / decomposition; As in; Usually these are unfurled (executed) in real time; So upon [N1] -> [N2] -> [N3] Is most often interrupted [N1] doesn’t become [N2] if it doesn’t satisfy [N2] doesn’t end in [N3] if the action isn’t verified Verification can happen upon multiple unfurling And multiple confirmation of such unfurling In this current iteration; We are experimenting with a scaffold length that looks like this $$$] $$] $] (for N1. N2. And N3) for reference ( we are currently experimenting with this scaffold type) $$$] LLLA LLLA LLLA each L is a slot for interruption / call $$] LLA LLA LLA (refer to Demo0.01 doc for this) $] LA LA LA Where the earlier the sequence is; the more Nodes there is to fill with Vision / Theme (refer to Demo0.01 entire agent build) injection nodes Whilst the closer it is to completion; the less space it is to inject /interfere ## Page 4 In biology (and in subsequent update) This may well be false / not exactly; (or may be exactly; should find out soon) But we presume that there [SHOULD] be a execution length variance Advanages may include . In order to both:  Later be easier to compress and detect on which completion  Later be easier to search and inject for specific "exact moment of Feel”  (mimicking biology in terms of motor execution)  i.e. (do it right after you see the guy twitches)  i.e. (do it right after the guy twitch; and you start charging)  i.e. (do it right after you almost connect the punch) if every thing is transcribed onto the $$$] $$] $] framework; everything is (likely) to be easy to detect on which sentence concept the nuance is referring to (and easily decomposed and injected between the communicating agents too) Imagine formatting every action – nuance to the same format of unfurling This will likely safe a lot of space / convention for multipurpose of the Same concept / reference (in the copy being used at the brain) ## Page 5 Some Example include Note that this is currently under works Decomposition picture An earlier thinking on unfurling Currently intended experiment (on the prototype agent) - - ## Page 6 A prior work describes it this way: BevCYC labeled EpiFoodPlau “designed to keep occurring to make a cat’s morning routine On scheduling and planning how to get food” “ has downstream Epi’s to then keep occurring with their respective completion” “ this is likely related to a Gene level in biology; first transcription becomes This BevCyc equivalent; second and third and subsequent translations Are also the same; either decomposed; copied; or learned and copied . “ food plau “ “ (bring out the template) “ (fill the template) “ (run several sim of the template) “ (put them on cycle* (refer to Page 12 For this) - Wake – roam – sleep - Roam – eat – roam v down (mechanism to breakdown / longer chain) - Roam – hunt – roam v down - Hunt – roam – ponder v down - Hunt – roam – seek v down (repetition noted here; its deliberate) - Roam – seek – ponder v down (functions as a data reconfirmation when rep) - Roam – hunt – roam (repeats) - ## Page 7 (IF)(hunt spots a prey)* (put hunting cycle below) (conclude both for reaching LIMB) - Roam – LIMB (ABC) - roam (leads to limb query)  (limb specs & limb tactics) other (IF)(energy). (IF)(priority (danger)). (IF)(self) (beacons.pdf for later) What is this cyle? these are ROS nodes (with their usual Checkpoint for completion) with some steps getting repeated to indicate how a typical chain must perform to make sure certain things (usually these are like so because of stamina / limitations on biology) - To just simplify it all 1. Make something really simple Wake- Safe- Sleep 2. Make a completion goal ROS style (exactly like ROS action steps) 3. Decompose each node until it reaches muscle tension % 4. If a creature is only a Single Cell + ProtoCilliate blob (feet) 5. Then Wake = chem. (checkpoint completion) Safe = decompose to where1 – move –where2 Move = is a muscle % . from that active decomposition Sleep = Move is accomplished Safe is accomplished. Sleep = trigger Chemical with Timer / Chemical completion token (somekind of mechanism for markers of well-rested) ## Page 8 How Complicated is this? Depending on how silly we decide Making a “Cat” with just tension globs (its just a proto-bacteria with some direction and muscle to sludge around) 6 / 8 / 12 BevCycles with 6 Decompositions Each (sounds like a 3 months project!) Not sure! I think that we don’t have to get it correctly though I think currently I can cook up a really silly cat that only has a Demonstration Of Social Cortex logging and applying (just recognition of Prey / Benefactor / Locations / Terrains / Patterns / Updates ) This is probably achievable Simply by having read-write nodes to the Demo0.01.doc (with an amateur team of 3/4/5) We can then revamp this for Proper Active inference modeling Upon more technical help! (note that these are described as “EXPERIMENTS”) Thus we are being liberal in proposing and :D Hypothesizing outcomes ## Page 9 Flexibility include Method to convert to more commonly known Frameworks of robotics BehaviorTrees (BT) . Hierarchical Task Networks (HTN) MotorPrimitives . STRIPS planning Onto robotics action - umm. This could be (hopefully) readily done - simply by adding a conversion node - and a conversion table - (for state / action division onto just actions ) Meanwhile for the checkpoints we *might be Able to emulate a different node To forward the completion trigger And that node could then be programmed / cached Seperately; however. That might cost a lot of Unoptimized process Mayhap we do an experiment Before moving on to Combination / revamps ## Page 10 Summary We intend to build the underlying framework (with CYC – Queue- Vision+Theme Decision Nodes Containers (mem) - and lots of transcribing agents) For a ROS-compatible cat (demo 0.01 doc) These are then a conceptually both the Sequence-Set; State-Instruction; and Action Instruction; Or atleast the same primitive on which They are all stored / re-copied for usage References “Along with advances and ideas posted around Youtube Documentaries and Interviews: Such as: (on bio-inspired agents) https://www.youtube.com/watch?v=8OhMmjlYvxU https://www.youtube.com/watch?v=6DrCq8Ry2cw https://www.youtube.com/watch?v=ZTE-JVd_QkA https://www.youtube.com/watch?v=L5bQnyq4OtQ https://www.youtube.com/watch?v=TXlfCY4m9jU .. many more --- *Extraction method: pymupdf*