Work led by @JasonJChoi1 @jasminejerrya Jingqi Li, Maria Gabriela Mendoza, Hamsa Balakrishnan and Claire Tomlin from @MITAeroAstro @MIT and @UCBerkeley @Berkeley_EECS #safety #collisionavoidance #marl #multiagentsystems #reinforcementlearning #CBFs

Our results, demonstrated through both simulations and physical drone experiments, show a substantial improvement in safety with minimal impact on operational efficiency.

By blending traditional control methods with multi agent reinforcement learning techniques, our approach helps them avoid getting stuck due to multiple constraints conflicting with each other or colliding, even in complex situations.

We propose a method that integrates layers of safety into multi-agent reinforcement learning algorithms to prevent collision avoidance.

#latepost Sharing our #RSS @RoboticsSciSys Conference #RSS2025 paper "Resolving Conflicting Constraints in Multi-Agent Reinforcement Learning with Layered Safety" joint work with @UCBerkeley & @MITAeroAstro Project page🔗 https://t.co/ktrOIVa1Ey Paper 📰 https://t.co/dWV4B3WYft

Come check out our paper at #ICRA2023! https://t.co/HkP2QI84yL @PatrikarJay

Code: https://t.co/YyzTDwifci Video:

Deploying robots in the real world poses the challenge of integrating real-life rules into AI policies. In this work, we explore encoding rules as high-level task specifications to improve the online performance of the LfD policies, using the example of aircraft navigation (2)

Thrilled to share that our work on combining Signal Temporal Logic (STL) and Learning-from-Demonstration (LfD) using Monte-Carlo Tree Search has been accepted to #ICRA2023! With @PatrikarJay, @Parvkpr, @JeanOhCmuBIG and @smash0190 at @AirLabCMU! 📜:

You can see more about our demo at the Intelligent Aerial Robotics here:

Our @ieee_ras_icra demo was featured on @robohub!

We have developed the first VFR AI pilot that enables autonomous aircraft to navigate a crowded airspace. The AI uses real world data collected at local airports to train vision, natural language and intent prediction modules that lead to safe and socially compliant navigation.

We are excited to announce the Tartan SLAM Series, an interactive series of presentations and discussions on SLAM! Open to both newcomers and experts. You can sign up on our website https://t.co/zjoOkuI30T to receive reminders and Zoom links to participate.

"Ingenuity has performed its first flight — the first flight of a powered aircraft on another planet!" The data reveals: Our #MarsHelicopter has had a successful first flight: 🚁

Wow!

Drop it like a bot. @NASAPersevere just released the debris shield that protected the #MarsHelicopter during landing. So begins the process to safely deliver Ingenuity to the surface of the planet. See all the steps here (p.20): https://t.co/Sr212FNyJO

Everyday, the drones in *my simulation* wake up and choose violence.

This is a pleasant surprise, #KRSSG featured by @OpenRoboticsOrg 🥳

Hacktober fest could revise their rules, specifically for India 👀

Planting a tree would be better than sending free T-shirts. Might make some people think twice before sending a pull request.🤔