Francisco Affonso

CS PhD Student @siebelschool | Robotics 🤖

[Francisco Affonso @faffonso2]

Thanks! The MoE actor learns its routing end-to-end, so there's no separate classifier or distillation step. On scaling, the experts specialize into shared sub-skills rather than whole tasks, so adding tasks doesn't mean adding experts. The critic is the bottleneck. We use one per task to isolate the signal, so critic count scales with the number of tasks. We're currently extending the framework to MTRL on loco-manipulation. Stay tuned!

[Francisco Affonso @faffonso2]

How do you get perceptive locomotion over rough terrain without brittle terrain classifiers?

Excited to share CTS-MoE, a framework for implicit terrain adaptation via Mixture-of-Experts for perceptive locomotion. No selectors or per-task policies; the policy adapts end-to-end straight from vision.

TL;DR:
→ Perception-driven routing handles diverse, discontinuous terrain implicitly; no high-level task selector or per-task policy distillation.
→ Big gains on hard tasks (climbing, gaps) under MTRL, with smooth transitions on both seen and unseen terrain.

🧵Thread:

[Robotics Papers @OWW]

CTS-MoE: Implicit Terrain Adaptation via Mixture-of-Experts for Perceptive Locomotion

Francisco Affonso, Matheus P. Angarola, Ana Luiza Mineiro, Aditya Potnis, Marcelo Becker, Girish Chowdhary
https://t.co/lJ2mxueVhj [𝚌𝚜.𝚁𝙾 𝚌𝚜.𝙰𝙸]

[Francisco Affonso @faffonso2]

[6/6] Huge thanks to my collaborators: @mpangarola, @aditya_potnis, Ana Luiza Mineiro, Prof. Marcelo Becker and Prof. Girish Chowdhary.

🌐 https://t.co/8tEG4vMOGd
📄 https://t.co/O8IZLbv46J

[Francisco Affonso @faffonso2]

[5/6] Hierarchical pipelines are bottlenecked by discrete high-level selection. Input-dependent routing lets the robot flow through transition states across a continuous 60m obstacle course.

[Robotics Papers @OWW]

CATNAV: Cached Vision-Language Traversability for Efficient Zero-Shot Robot Navigation

Aditya Potnis, Francisco Affonso, Shreya Gummadi, Naveen Kumar Uppalapati, Girish Chowdhary
https://t.co/vplnQSBwpf [𝚌𝚜.𝚁𝙾]

[Robotics Papers @OWW]

Learning Terrain-Specialized Policies for Adaptive Locomotion in Challenging Environments. https://t.co/QQm24Pld6L