paper-conference

We leverage GPUs to create probabilistically collision-free convex sets in configuration space on-the-fly, enabling high-quality trajectory optimization without challenging collision avoidance constraints. Our approach runs 17× faster with 28% higher reliability than traditional methods, validated on both simulations and a KUKA robot with real-time perception.
[RSS, 2025]

We propose two new algorithms for constructing collision-free convex polytopes in robot configuration space. Our key insight is that finding near-by configuration-space obstacles using sampling is inexpensive and greatly accelerates region generation. Our two algorithms use such samples to either employ nonlinear programming more efficiently (IRIS-NP2 ) or circumvent it altogether using a massively-parallel zero-order optimization strategy (IRIS-ZO). We show that IRIS-ZO achieves an order-of-magnitude speed advantage over its predecessor IRIS-NP. IRISNP2, also significantly faster than IRIS-NP, builds larger polytopes using fewer hyperplanes, enabling faster downstream computation.
[ISRR, 2024]

We propose an algorithmic appraoch to approximating robot configuration space with a small collection of convex sets. First, the algorithm constructs a visibility graph using sampling, then a small clique cover of the visibility graph is computed, and finally the cliques are inflated to full-dimensional polytopes that are collision free.
[ICRA, 2024]

We present an efficient method for certifying noncollision for piecewise-polynomial motion plans in algebraic reparametrizations of configuration space. Such motion plans include those generated by popular randomized methods including RRTs and PRMs, as well as those generated by many methods in trajectory optimization. Based on Sums-of-Squares optimization, our method provides exact, rigorous certificates of non-collision; it can never falsely claim that a motion plan containing collisions is collision-free.
[ICRA, 2024]

Autonomous racing is a challenging task that requires vehicle handling at the dynamic limits of friction. While single-agent scenarios like Time Trials are solved competitively with classical model-based or model-free feedback control, multi-agent wheel-to-wheel racing poses several challenges including planning over unknown opponent intentions as well as negotiating interactions under dynamic constraints. In this work, we address these challenges via self-play reinforcement learning to enable zero-shot sim-to-real transfer of highly dynamic policies.
[CORL, 2023 + Best Paper Award ICRA Multi-Robot Learning Workshop, 2023]

We show that action discretization and value decomposition, enables us to scale and apply Q-learning to continuous control tasks. With bang-bang actions alone, the performance of this critic-only approach matches state-of-the-art continuous actor-critic methods when learning from features or pixels.
[ICLR, 2023]

In this work, we find and certify collision-free polytopes in a rational parametrization of robot configuration space using convex optimization.
[Best Paper Award WAFR, 2022]

Cameras enable integrated proprioceptive sensing of soft bellows actuators.
[IROS, 2020]