Bounce and Learn: Modeling Scene Dynamics with Real-World Bounces
Authors: Senthil Purushwalkam, Abhinav Gupta, Danny Kaufman, Bryan Russell
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To achieve our results, we introduce the Bounce Dataset comprising 5K RGB-D videos of bouncing trajectories of a foam ball to probe surfaces of varying shapes and materials in everyday scenes including homes and offices. Our proposed model learns from our collected dataset of real-world bounces and is bootstrapped with additional information from simple physics simulations. We show on our newly collected dataset that our model out-performs baselines, including trajectory fitting with Newtonian physics, in predicting post-bounce trajectories and inferring physical properties of a scene. |
| Researcher Affiliation | Collaboration | Senthil Purushwalkam & Abhinav Gupta Robotics Institute, Carnegie Mellon University {spurushw,abhinavg}@cs.cmu.edu Danny Kaufman & Bryan Russell Adobe Research {kaufman,brussell}@adobe.com |
| Pseudocode | No | The paper describes the model architecture and training procedures in detail using text and diagrams (Figure 2, 6, 7), but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Dataset and code available here: http:// www.cs.cmu.edu/ spurushw/projects/bouncelearn.html |
| Open Datasets | Yes | To train our model, we introduce a large-scale Bounce Dataset of 5K bouncing trajectories of a probe with different surfaces of varying shape and material in everyday scenes including homes and offices. Dataset and code available here: http:// www.cs.cmu.edu/ spurushw/projects/bouncelearn.html |
| Dataset Splits | Yes | For the forward-prediction task, we split our dataset into training, validation, and test sets containing 4503, 196, and 473 trajectories, respectively. |
| Hardware Specification | No | The paper does not specify the hardware (e.g., GPU models, CPU types, memory) used for running the experiments or training the models. |
| Software Dependencies | No | The paper mentions Py Bullet Physics Engine (Coumans & Bai, 2016 2017) and uses architectures like AlexNet and PointNet, but it does not provide specific version numbers for software libraries, frameworks (like TensorFlow or PyTorch), or other dependencies. |
| Experiment Setup | Yes | We update the parameters of the PIM using a batchsize of 32, initial learning rate of 0.01 and weight decay of 0.0005. The learning rate is dropped by a factor of 10 after every 32000 iterations. The training is done for a total of 96000 iterations. We update the parameters of the PIM and VIM using a batchsize of 32, initial learning rate of 0.001 and weight decay of 0.0005. The learning rate is dropped by a factor of 10 after every 8000 iterations. We observe that the training converges at 24000 iterations. |