The process is explained here : [ Ссылка ]

The other levels of Tomb Raider 1 are here : [ Ссылка ]

----------
New Feature:

The bot now has a third memory system that can store chronological events. When it gets lost (detecting a loop or exhausting all possible textures), this memory system can be triggered. It captures the identified words from these textures that as a link with Lara's personality. When activated, the bot will compare the words from the last event to previous ones, using synonyms to establish matches. If a match is found, the bot will know its place in the chronology and attempt to identify the texture from the previous step until it reaches the matching one.

Example in Practice:

1) The bot identifies a lock, it still store it in the long term memory because it has a link with Lara's personality traits, then it stores the words linked to this texture it in the new third memory, this is saved as step 1.

Note : this is important to make it uses synonyms because it increases the probability of matching later, depending of moment some identification could get different results, a fence could be a barrier depending of the angle.

2) It continues searching for textures and encounters a bridge, storing its words in the third memory.
3) The third memory now contains: 1: lock, synonym, 2: bridge, synonym.
4) The bot finds a key and adds it to the third memory, so : 3: key, synonym.
5) When the bot is stuck, it activates the third memory, it knows the last step was the step 3, so it looks for the step 2: bridge, synonym.
6) Once he find a new texture, it scans it and also makes synonyms, then try to find a match.
7) When the match is found it unlocks the next step, now it looks for the step 1: key, synonym.
8) When it reaches the first step it put the key on the lock.

Problems:

In some cases, level design makes it impossible to backtrack. Therefore, using the key should be a "last resort" when all routes are identified, as if the bot spots new textures during the third memory process, it will follow them, potentially causing unnecessary actions. It's crucial to allow the regular identification process to continue.

The captured textures used for the steps aren't always identifiable from a distance, so I've set the render distance to the maximum. Even if a human could identify a single black pixel as a lock, the bot may still search when it doesn't seem necessary. Improving accuracy would require a 4K screen instead of the current Full HD one. Any clipping would pose a significant challenge for this bot.

New Features:

If the bot identifies textures with text it will try to read it. This is included in google search. For best results this identification will happen when the texture is large enough on the screen. This doesn't work in the inventory yet.

Items, such as Scions, are now separated from keys. I won't try to insert it anywhere.

When the bot must select an item, the rotation of the item manager is no longer random. It will rotate to the closest direction to reach the needed item. However, to maintain realism, there's still a 10% chance of selecting the wrong direction.

Fix:

When rotating Lara's body, the bot will now favor turning to the right more frequently.

Future Update:

When there are multiple locks, the bot can't determine where to use the key. To address this, it will check every key. Humans associate the color of the lock with the color of the key, but the bot can't do this yet. The idea is to blur the key to obtain its average color and match it with the average color of the lock. This method should work for most items, as the game uses the same texture for keys and locks, except for the Egypt items, which rely on their shape.

Information:

This isn't a live result; it's explained in the link above. The game is paused every time a comment is generated. The result presented here combines all the parts, which are numerous. Therefore, you can't have a normal conversation with Lara Croft because each comment takes many minutes to generate.

Lara is still an AI trying to create the illusion of awareness. This AI appears convincing because I instructed it to consider Lara's personality, which it knows based on the information I provided about Lara Croft. However, if you focus solely on the comments, you'll notice it's still a robot. Everything it says or does can be explained.

This video demonstrates the potential of combining multiple free AIs to create something. It serves as an example to showcase the capabilities of these technologies. Of course, I didn't create these technologies; they are available for anyone to try.

While this demo is impressive, our human brains remain more efficient, and having a real-life conversation is more remarkable. We should appreciate our self-awareness and use AI for positive purposes.

свернуть