Learning to Throw Objects Safely in Multi-Obstacle Environments
Mohammadreza Kasaei, Klemen Voncina, Hamidreza Kasaei
AI summary
Problem
Reliable robotic throwing in cluttered environments remains underexplored, as existing methods assume obstacle-free settings or fail to scale with increasing obstacles.
Approach
The authors introduce a potential field state representation that encodes basket attraction and obstacle repulsion on a fixed grid, bootstrapped by kinesthetic teaching and optimized via reinforcement learning.
Key results
- SAC outperforms DDPG and TD3 in consistent throwing performance
- Potential field representation scales to arbitrary obstacle counts unlike explicit pose encodings
- Achieves up to 90% success rate in real-robot cluttered scenes
- Policies generalize effectively to unseen throwable objects and obstacle configurations
Why it matters
Enables high-throughput, collision-free object placement for logistics and automation systems operating in unstructured environments.
Abstract
Robotic throwing enables fast and efficient ob- ject placement beyond the robot’s immediate workspace, but reliable throwing in cluttered environments remains underex- plored. Existing approaches, such as TossingBot, learn throwing strategies from visual input but assume obstacle-free settings. In this paper, we address the problem of throwing objects into a target basket while avoiding obstacles placed randomly in the scene. We introduce a potential field state representation that compactly encodes both basket attraction and obstacle repulsion on a fixed-size grid, enabling reinforcement learning (RL) policies to generalize across arbitrary numbers and config- urations of obstacles. The policy is initialized from kinesthetic demonstrations and optimized in simulation using three state- of-the-art RL algorithms (SAC, DDPG, TD3). Among these, SAC achieves the most consistent performance across scenarios. We compare the potential field representation against explicit state encodings and demonstrate that it achieves higher success rates and better scalability to unseen obstacle configurations. Real-robot experiments with unseen throwable objects confirm robust sim-to-real transfer, achieving up to 90% success in cluttered scenes. These results demonstrate that PFR provides a practical and robust representation for safe and efficient robotic throwing in unstructured environments. A video showcasing our experiments has been attached to the paper as supplementary material.