Week 3 Our 6 ideas

This week, we mainly focused on ideation, branding, and preparation for the quarter presentation.

Our excellent designer prototyped several ideas for our project’s poster and logo, and finally, we decided on the following designs:

Team logo
Team Poster

We tried to present the information that the computer could act like a human being in video games. Even though we might modify our branding more after the Quarters, our current designs are good enough to present to our visitors next week.

Besides the branding, on Wednesday, our team did a short trip to Google headquarter in Mountain View for the team photo and decoration supplies. We took a very lovely photo in front of the google sign in the convenience store:

Team photo

For the preparation of the Quarter, we planned to have four different parts of presentation: introduction, tech demo, design process, and design ideas. We will lead our guests to the introduction section first, by explaining what our project is about and what is reinforcement learning, and the problem we are trying to solve. Secondly, we will show our tech demo, which is a small prototype about a ball and a platform. The platform is the AI, moving against the ball to prevent it from falling down. By showing this demo, we tried to explain that the existing Unity Machine Learning Plugin works for reinforcement learning. Lastly, we want to present the flow chart of our design process and our current stage. We will explain our current stage more by listing our six design ideas.

The six ideas come from every teammate’s effort. We researched the potential applications of enhancing the relationship and interaction and came up with six ideas:

  1. Party game – Mario Party/Overcooked

In Mario Party mini-game, the player is matched with NPC who might not have the same abilities as the player, or like Overcooked, the player could not play without friends. The problem we are trying to solve is that right now, NPC cannot have the same ability as other teammates and this leads to an unbalanced team. What if the NPC could have the same ability as the player?Through creating human-like agents, the goal is not to have a smart agent completing tasks accurate and fast for you, instead, to create agent who shares the same ability as other human players so the teamwork could be more balanced. Because we cannot do real-time training. We will collect data from the human players, train NPCs to act more like a human in the next version. Another aspect is to think the agent has different abilities, so maybe there is clumsy or dictated chef in the game.

2. Sims- Raise up a kid

This idea comes from the Sims game. The player is the parent has a kid Sim to raise. They will only give choices to the kid (the NPC), without telling him/her what is right or what is wrong. The kid, on the other hand, will get their parent’s emotion feedback and make choices. The problem we are trying to solve is to build up the emotion between the player and the NPC throughout the gameplay. Better than a state machine, game designers don’t have to create all the scenarios and edge cases. RL will invent them.

3. Responsive Canvas

This idea comes from the game we played in the improv class. The player and the NPC take turns to finish an artwork on a blank canvas. The agent will either add on or erase strokes/colors after the player’s turn. The player will learn how to paint like a master as Van Gogh or Picasso, with the NPC’s guidance. This idea is very innovative. However, it might meet problems with imaging processing and edge detection technology.

4. Escape from the fire scene

A couple is escaping from a building on fire. One of them is the leader, and the other is the follower.  The Leader decides the way to run and avoid getting burned within 3-min. The Follower tries to open doors, save people, or grab objects to earn scores. Both of them consume Energy. Whoever has the highest Energy in real-time will be the Leader. The conflict scenario we are trying to develop is when trying to escape within a time limit, follower decides to save a baby, but the other wants to leave the scene asap. The uniqueness of this is that RL might respond differently according to different situations, i.e., the intimacy of the couple, and create tension between the player and the NPC companion.

5. Plants war

We want to create a situation that there is a neutral party in the game. The game will have two NPCs. Plus the player, there will be three parties in the game. The player is vegetation fighting animals. The animal is the enemy and the RL agent is the bird. The Player tries to destroy the animal’s base. During the battle, the bird comes out to find food. The player can use bird as decoy or shield. When the Agent needs food, it will try to come out to find the food. During the process, the Agent will interact with the player or other NPCs, it will get rewards (points) for the interaction.

6. Co-op puzzle game

The last idea is a co-op puzzle game. In the existing puzzle game, there is limited guidance from the NPC character to solve puzzles. There are only little interaction and character depth. We want to improve the interactions between the player and the NPC using RL. We have an RL-trained companion NPC character who can provide help or solve puzzles with the player more dynamically according to the player’s behavior. It allows real-time feedback between the player and the agent using limited communication such as physical movements and emoji stickers. The player figures out how to help the agent solve a puzzle in order to complete a quest. The agent’s goal is to maximize the player’s experience through guiding and responding in the most effective way according to player behavior and the puzzle-solving progress. They will need to rely on some level of communication to solve the puzzle.     

Overall, we hope that we could get feedback from the guests about our six ideas and move forward to design our first chosen idea soon after the Quarters. We are looking forward to the Quarters next week!