AndrejOrsula / drl_grasping

Deep Reinforcement Learning for Robotic Grasping from Octrees

https://arxiv.org/pdf/2208.00818

AndrejOrsula/drl_grasping Issues

Random object - Make sure models belong to the Fuel collection if sampled from local cache
Closed 5 months ago
Replace fork of Gym-Ignition with upstream release (LocalCache)
Closed 5 months ago1
Deploying on Franka Research 3 (real)
Closed 9 months ago2
Training Error & Docker Tips
Closed a year ago6
How much SDRAM is needed to reproduce the results in paper?
Closed a year ago
Gazebo Source Installation
Closed a year ago6
Pyoctree Attribute Error
Closed a year ago2
Caffe Installation
Closed a year ago1
Error while running training bash script
Closed 2 years ago2
Kinova arm with SAC Algorithm not learning
Closed 2 years ago4
Refactor drl_grasping module to simplify usage and modifications
Closed 2 years ago1
EPIC: Future works
Closed 2 years ago1
Visualizing Reach Tasks
Closed 2 years ago2
Adding static Objects to the Reach Task
Closed 2 years ago2
Error running ex_train.bash in a headless way
Closed 2 years ago3
`ex_enjoy_pretrained_agent.bash` hanging in fresh Docker image
Closed 3 years ago4
`ocnn` dependency cannot be built using `colcon`
Closed 3 years ago2
Add instructions for examples in README
Closed 3 years ago2
O-CNN repository for local installation
Closed 3 years ago11
Refactor adding of controller inside panda model
Closed 3 years ago1
Refactor gripper handling inside the task
Closed 3 years ago2
Add few trained agents that can be enjoyed
Closed 3 years ago1
Move all animations from README to another repository and purge them from this repo's history
Closed 3 years ago1
EPIC: Design and decisions
Closed 3 years ago1
Extend 3D observations with proprioceptive data
Closed 3 years ago1
Make sure random objects don't get spawned inside each other, while getting spawned as close to ground plane as possible
Closed 3 years ago2
Move tf broadcasting from randomizer into task, while making it camera/robot agnostic
Closed 3 years ago1
Curriculum learning for grasping environment
Closed 3 years ago1
EPIC: Reward function for grasping environment
Closed 3 years ago3
Add additional terminations that result in failure
Closed 3 years ago2
Reset of joint positions causes segmentation fault
Closed 3 years ago1
Implement a custom replay buffer for octree observations
Closed 3 years ago1
Experiment with low-poly version of panda (collision mesh) to see if it brings any performance for training
Closed 3 years ago1
Add dockerfile
Closed 3 years ago1
EPIC: Hyperparameter tuning
Closed 3 years ago1
EPIC: 3D observations and policy
Closed 3 years ago1
Estimate normals from point cloud
Closed 3 years ago1
Determine what kind of reward function is descriptive enough (with preference for sparse reward)
Closed 3 years ago1
Enable control of full 3D orientation
Closed 3 years ago1
Try to use "6D representation" for orientation representation
Closed 3 years ago1
Implement observations/policy for 3D spatial (e.g. occupancy) data
Closed 3 years ago1
Convert point cloud to octree, using normal vectors as features
Closed 3 years ago1
Extend 3D observations/policy with visual (RGB) data
Closed 3 years ago1
Use a shared feature extractor between policy and q-network(s)
Closed 3 years ago1
Eliminate the need to run unpaused steps during each reset
Closed 3 years ago1
Make sure a new observation is received after each reset (Ignition sensors)
Closed 3 years ago1
Eliminate the need of having base worlds for tasks (use default world instead)
Closed 3 years ago1
Autotune hyperparameters
Closed 3 years ago1
Tensorboard is currently not creating logs due to protobuf version discrepancy between Gazebo and Tensorflow
Closed 3 years ago1
Decide what kind of 3D data representation and corresponding network architecture to use for observations/policy
Closed 4 years ago1