Skip to content

Commit

Permalink
update posts
Browse files Browse the repository at this point in the history
  • Loading branch information
MoeBuTa committed May 7, 2024
1 parent eee1fc5 commit 3f8239a
Show file tree
Hide file tree
Showing 4 changed files with 57 additions and 10 deletions.
67 changes: 57 additions & 10 deletions _posts/2024/05/2024-05-06-progress-meeting-notes.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,16 +6,11 @@ imgUrl: /assets/source/image/blog/
---


## Literature reviews

[Overleaf](https://www.overleaf.com/project/660fbf6b9ea6dbdd103450d1)

## Asynchronous LLM Request / Control
- thread 1 running rule-based safety checks
- thread 2 running LLM control
- thread 3 running user query


```mermaid
graph TD;
Expand All @@ -28,21 +23,73 @@ graph TD;
C --> |Control|R
R --> |Update|A
R --> |Trigger|D
```

![img]({{page.imgUrl}}eyesim-seq.jpg)


## Data for Retrieval

- scene graph

![img]({{page.imgUrl}}scene-graph.png)

- sensor-action graph (May constructed from hard coded script)
- giving a set of sensor data, find k most similar sensor nodes.
- may start with one sensor with a few range of values
- e.g.: distance sensor, range [0, 1000], 0-200, 200-400, 400-600, 600-800, 800-1000, >1000
- query these sensor nodes, list all connected action nodes with priorities, send to LLM

```mermaid
graph TD;
A[Sensor Data 1] --> B(Action1)
A --> C(Action2)
A --> D(Action3)
E[Sensor Data 2] --> F(Action4)
E --> G(Action5)
E --> B
H[Sensor Data 3] --> C
C --> G
H --> A
```


- trace logs

- map


## Lawnmower, Maze

hard coded script vs zero-shot LLM vs LLM + RAG

## Issues

- Prompt, retrieve content
- Task scenarios brainstorming (May start with finding objects in the environment)
- ~~have all the information of the environment (may not be so interesting)~~
- ~~e.g.: the can is at (x,y), go and find it~~
- ~~retrieve map, scene graph, sensor-action graph, trace logs.~~
- have limited information of the environment
- e.g.: there is a can behind the wall, try to find it
- retrieve scene graph, sensor-action graph, trace logs.
- no information of the environment
- e.g.: try to find a can somewhere in the world
- retrieve sensor-action graph, trace logs.
- have incorrect information of the environment
- e.g.: the can is at (x,y), but it is not there
- retrieve map, scene graph, sensor-action graph, trace logs.
- dynamic environment
- e.g.: the can is moving, go and find it
- ~~Prompt, retrieve content~~
- LLM response parsing
- response content
- response format
- User query


## TODOs:

1. Grab data from hard coded script
2. Construct Graphs
3. Prototype
Binary file added assets/source/image/blog/eyesim-seq.jpg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file modified assets/source/image/blog/rag-arch.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added assets/source/image/blog/scene-graph.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.

0 comments on commit 3f8239a

Please sign in to comment.