EQA merge & DynaMem CPU#254
Conversation
…/stretch_ai into hello-peiqi/grapheqa_merge
|
|
||
|
|
||
| class SparseVoxelMap(SparseVoxelMapBase): | ||
| def __init__( |
There was a problem hiding this comment.
It would be good to add a docstring here, this is a lot of params (mostly borrowed - can use @OVERRIDES)
| DYNAMEM_VISUAL_GROUNDING_PROMPT, model="gpt-4o-2024-05-13" | ||
| ) | ||
|
|
||
| def find_alignment_over_model(self, queries: str): |
| print("Points close the the point are not similar to the text!") | ||
| return torch.max(alignments[distances < distance_threshold]) >= similarity_threshold | ||
|
|
||
| def get_2d_map( |
There was a problem hiding this comment.
Is this different from the base one? If not - just override
if so maybe we could do some refactoring for less duplication?
| if debug: | ||
| import matplotlib.pyplot as plt | ||
|
|
||
| plt.subplot(2, 2, 1) |
There was a problem hiding this comment.
you can probably delete this debug code
| print(f"Error: {e}") | ||
| return None | ||
|
|
||
| def localize_with_mllm(self, text: str, debug=True, return_debug=False): |
There was a problem hiding this comment.
this new function could probably use some documentation
| rr.init("Stretch_robot", spawn=spawn_gui) | ||
|
|
||
| if output_path is not None: | ||
| rr.save(output_path / "rerun_log.rrd") |
There was a problem hiding this comment.
do you find these rerun logs useful?
There was a problem hiding this comment.
Not yet, but I guess it is for automatically saving rerun res; since grapheqa does that, I think it is useful especially for some researchers want to collect demo for their websites.
| # hellorobotinc/stretch-ai_cuda-11.8:$VERSION: the docker image to run | ||
| run_docker_command run \ | ||
| -it \ | ||
| --runtime nvidia \ |
There was a problem hiding this comment.
why do you get rid of this? is it no longer necessary?
There was a problem hiding this comment.
It gives me bugs
There was a problem hiding this comment.
Can you still run with access to the NVIDIA hardware?
| self._manipulation_radius = parameters["motion_planner"]["goals"]["manipulation_radius"] | ||
| self._voxel_size = parameters["voxel_size"] | ||
|
|
||
| # self.image_processor = VoxelMapImageProcessor( |
|
|
||
| def execute_action( | ||
| self, | ||
| text: str, |
There was a problem hiding this comment.
documentation - say what actions can be?
* update? * clean up codes * update docs * update * update * cpu * update * update * change models * fix bug * installation * debug * tune parameters * restore gui * fix bug * update docs and args * fix bug
* update a bit * small update * revert some changes * revert some changes * revert some changes * automatically decide which model to use --------- Co-authored-by: GRAIL <grail@demobox.hydra-bee.ts.net>
* remove moonlight * remove dinobot * remove yolo * remove detic * update installation * remove the remains of Detic * remove paligemma * remove molmo * Hello peiqi/remove online audio (#278) * remove audio * tune audio speed * add piper * add new engine * modify piper * modify piper * fix bugs * update test * update test * update tets * rearrange files * small update * small update * remove cli * remove more files * remove voice chat app * remove useless files * remove useless folder * add files * new gitignore * fix bug * update gemma and docker * add terminate on failure * Hello peiqi/clean app (#279) * remove files * remove emote app * update keyboard teleop * update print states * merge check status and print states * merge check status and print states * remove start rerun * remove more perception modules * remove vlm planning * replace git with blip * remove dinov2siglip * restore scripts * remove mobile sam * deal with sam * update docs * update install.sh * remove map loader app * remove more apps * remove load map util * remove demo * remove ENV and Agent * clean core * clean drivers * update detic * unify agent * clean robot agent * remove get detections * remove unused attr * fix bug * fix ai pick up bug * Hello peiqi/fix simulation (#283) * fix * fix intrisincs * fix not robocasa parts * update * tune readme * Hello peiqi/slam checking (#281) * remove readme * remove unnecessary core * remove files * restore * remove navigation utils * remove navigation app * remove drivers * fix orbslam3 bugs * fix orbslam3 bugs * remove log * remove unnecessary ros2 files * further clean * update slam * checkpoint * remove orbslam docs * update detic * clean more files * clean slam * update * fix bug * fix typo * fix match method * fix bug * installation * debug grasp * slight tunning
hello-robot-shehab
left a comment
There was a problem hiding this comment.
I was able to run the apps in the PR on my Stretch 3.
Anyone want to test this system based on this instruction Codes are currently on hello-peiqi/grapheqa_merge branch.
You don't have to worry about discord bot.
I also modified DynaMem and AI Pick up. So please also check:
Some testing steps recommended:
./install.shpython -m stretch.app.run_eqa --robot_ip $ROBOT_IP(You will need Gemini token)python -m stretch.app.run_dynamem --robot_ip $ROBOT_IP --visual-servo -M(You will need OpenAI token)./install.sh --cpu. Then test DynaMem on CPUpython -m stretch.app.run_dynamem --robot_ip $ROBOT_IP --cpu --match-method "class(dynamem)" --vspython -m stretch.app.run_eqa --robot_ip $ROBOT_IP -DStart with some simple questions such as "is there a monitor on the table?" "what is the color of the chair"